OpenCV - OpenCV DevZone

1

OpenCV Meeting notes for 2009 year

2

1

3

1

{{>toc}}

4

1

5

1

h2. 2009-01-06

6

1

7

1

*_Agenda_*

8

1

9

1

 * Update

10

1

 * VTK?

11

1

 * Finding planes with Radu's code?

12

1

13

1

*_Minutes_*

14

1

15

1

 * Russian holidays

16

1

 ** *Vadim* working on using Lapack+BLAS in OpenCV

17

1

 *** Using ATLAS & MKL

18

1

 * See about getting some of *Radu* Ruso's mesh routines in OpenCV

19

1

 * Want to move C-descriptor into OpenCV

20

1

 *** What's needed to make a paper out of it

21

1

 * 3D reconstruction

22

1

 ** Good camera

23

1

 ** Textured light

24

1

 ** Rotating platform and rotate object captured frame by frame

25

1

 ** Robot can scan object

26

1

 * Structure from motion

27

1

 ** What is JD working on with 3D reconstruction

28

1

 * "OpenCV" entry in manufacturing robots

29

1

30

1

31

1

*_Action Items_*

32

1

33

1

*Gary*

34

1

35

1

 * Talk to JD about code

36

1

 * Record minutes of tomorrows discussion with the Vision workbench team

37

1

 * Talk to Calendar about producing a Star-C-descriptor paper

38

1

39

1

*Vadim*

40

1

41

1

 * Compare OpenCV star detector with ROS star detector

42

1

43

1

 h2. 2009-01-13

44

1

45

1

 * Updates

46

1

 * Interface issues (what next)

47

1

 * Pattern Recognition

48

1

 * 3D functionality

49

1

50

1

*_Minutes_*

51

1

52

1

Update *Gary*:

53

1

54

1

 * Set up stereo capture node and will collect 100's of outlets and doorknobs

55

1

 * Collecting 3D object database of Ikea objects

56

1

 * Writing an opencv tutorial for Stanford class, will post that

57

1

58

1

Update *Vadim*:

59

1

60

1

 * Continued work on BLAS and LAPACK interface

61

1

 ** Eigen vectors and values,

62

1

 ** SVD,

63

1

 ** cvsolve,

64

1

 ** cvinvert and

65

1

 ** qr decomposition

66

1

 ** Accuracy is great and performance is great

67

1

 *** 500x500 matrix is 2.5x faster

68

1

 *** Small matrixes, 10x10, the new implementation is slower

69

1

 **** Need to add branches in this case

70

1

 ** Netlib.org Jack Dungar's implementation, puts out clapack with same interface

71

1

 *** Eigen2 lib, small matrices are faster than lapack, but algorithms aren't as well developed as lapack

72

1

 * multiwin replaced using wImage (code from Google)

73

1

 * This week is after Russian new year vacation.

74

1

 ** Anatoli working on Octrees for Spin images

75

1

 * CVS becomes obsolete this week

76

1

77

1

Update *Victor*

78

1

79

1

 * Was sick with mono for 3 weeks

80

1

 * Outlets

81

1

 ** Filter keypoints using random forest

82

1

 *** Train image passes around keypoints

83

1

 *** 2 class filter for inside/outside outlets, works 90%

84

1

 *** 70% recognition on hard datasets with textured walls and different lighting condition

85

1

 *** Low false positive rate

86

1

 **** Need to turn down false positive rate to gain low false negative (missed detection) rate and then apply stronger models to the detections

87

1

 *** Needs to wait to final camera setup

88

1

 ** Need datasets for US outlets for keypoint detection

89

1

 *** Adapt hole searching code for US outlets

90

1

 * Doorknobs

91

1

 ** Dark round objects using series of thresholds and connected components

92

1

 ** RCG collected doorknobs together with chessboards.

93

1

 *** Perspective corrected

94

1

 *** Trained adaboost

95

1

 *** Same for test set

96

1

 **** Results weren't that good -- 40% detection with 15% false alarm

97

1

 **** *Gary* suspects making many different templates to match -- one each 10 degrees.

98

1

 * Random forest improvements

99

1

 ** Extreme random trees code implemented

100

1

 *** Bugs fixed while in hospital

101

1

 *** Compare with same datasets measure test errors.

102

1

 *** Errors with noise variables

103

1

 *** Variable importance -- correction noise variables

104

1

 *** If improvements, possible paper

105

1

106

1

Videri camera is supported in OpenCV

107

1

108

1

Nearest Neighbor

109

1

110

1

 * cvFeatureTree, already in OpenCV

111

1

 * David Lowes FLANN -- test against cvFeatureTree()

112

1

113

1

*_Action Items_*

114

1

115

1

*Gary*

116

1

117

1

 * Get doorknob and outlet datasets to *Victor*

118

1

 * Get outlets from Milestone dataset to *Victor*

119

1

 * Get cannon outlet sites to Argus or Willow sites

120

1

 * Get Willow doorknobs to *Victor*

121

1

122

1

*Vadim*

123

1

124

1

 * Send SVN instructions for Window to *Gary*

125

1

126

1

*Victor*

127

1

128

1

 * Test random trees against random forest

129

1

130

1

h2. 2009-01-20

131

1

132

1

*_Agenda_*

133

1

134

1

Update:

135

1

136

1

 * Milestone2

137

1

 ** Outlets up close

138

1

 ** => *Victor* to Doorknobs?

139

1

 *** 3D box with chessboard

140

1

 * Planning for Summer release, what should be in it

141

1

142

1

*_Minutes_*

143

1

144

1

*Gary*

145

1

 Close range outlet detection to subpixel

146

1

147

1

148

1

*Vadim*

149

1

150

1

# LAPACK is now included into OpenCV tree.

151

1

cvSVD, cvEigenVV, cvDet, cvSolve and cvInvert have been rewritten using LAPACK;

152

1

cvInvert and cvSolve have got an additional method CV_CHOLESKY, cvSolve also got additional method CV_QR.

153

1

# *Anatoly* has implemented a draft version of octree for computing spin images. The existing implementation of octree did not contain the key algorithm for retrieving all the points in the cloud that belong to the given sphere and it was difficult to integrate it, so the new class has been written. *Anatoly* compared performance of this new code with OBB tree from VTK, which has been used by the existing spin image code and found that the performance of the two is very close now:

154

1

155

1

 * Hope to speed up spin feature by factor of 10

156

1

 * Future to speed up Radu's histogram of normals

157

1

158

1

Next release

159

1

160

1

 * C++ interface with templates

161

1

 ** 30-50% done

162

1

 ** Templates will be optional

163

1

 ** More compatible with STL

164

1

 * Much better python interface

165

1

 ** Mark submitted a patch for python wrapper and use with numpy

166

1

 * 3D features/related algorithm

167

1

 ** Spin images -- goal is 10x speedup.

168

1

 ** Plane fitting from Radu

169

1

 * Alpha and/or 4 channel support

170

1

 * 2D features

171

1

 ** C-descriptor (Calandor descriptor)

172

1

 ** Maybe Boundary fragments

173

1

 * Stitching/VSLAM

174

1

 ** Move in the V-SLAM stitching into OpenCV (?, more probably next ...)

175

1

 ** Place localization/recognition as an app or example code for C-Descriptors

176

1

 * Optical flow

177

1

 ** Dense

178

1

 ** Rigid

179

1

 ** Affine

180

1

 ** Intern from Michael Black

181

1

 * Machine learning library

182

1

 ** Extremely Randomized Trees

183

1

 ** Nistor Tree

184

1

 ** Convolutional NN

185

1

 ** Additions to RF

186

1

 ** Extend cascade to other features beyond Haar

187

1

188

1

189

1

*_PRIORITIES_*

190

1

191

1

 # C++ interface

192

1

 # Python

193

1

 # 3D Features (for Obj rec)

194

1

 # 2D Features (for Obj rec)

195

1

196

1

(no order)

197

1

198

1

 * MLL

199

1

 * Alpha / multi-channel

200

1

 * Optical flow improvements

201

1

 * Stitching

202

1

203

1

204

1

*Victor*

205

1

Extremely Randomized Trees (ET) algorithm shows comparable performance with published in the original Geurts paper on 4 datasets while on 2 datasets (spambase and vehicle) it demonstrates larger errors (see the table below). We are still investigating this issue, the current main hypothesis is that there is a bug that plays a significant role only on these two datasets. Variable importance charts built for Random Forest (RF) and ET models are similar.

206

1

207

1

208

1

 | Name | RF MLL | ET MLL |	ET Geurts |

209

1

 | waveform |	17.1 |	16.06 |	16.61|

210

1

 | vehicle |	21.88 |	47.29 |	26|

211

1

 | spambase |	5.41 |	22.68 |	4.17|

212

1

 | isolet | 6.61 |	10.52 |	7.61|

213

1

 | twonorm7400 |	4.523944 |	3.745071| 3.53|

214

1

 | ringnorm7400 |	6.164789 |	3.138028 |	3.27|

215

1

216

1

There is a bug on Vehicle and Spambase.  *_UPDATE:_* This bug is fixed.

217

1

218

1

The application for automatic tracking of planar objects has been tested on windows and linux with usb camera and video.

219

1

220

1

A method for learning an RF model of an outlet invariant to affine transformations has been implemented. An RF is learned on patches of frontal view outlets that have been distorted with random affine transforms (following Lepetit's paper). Each outlet candidate gets a weight (in [0,1]) from this model and low weighted samples are filtered out. Attached is a precision-recall curve of this algorithm obtained by varying the weights threshold. It shows the performance of the model (recall = 1 corresponds to no filtering). The most of false positives are gathered on few images with rich texture.

221

1

222

1

 * Have outlets detectors in 2D

223

1

 ** Single image

224

1

 ** At least 50x50 ROI

225

1

 ** Based on finding dark round objects (European) and dark round (American)

226

1

 ** 2 stage RF (European):

227

1

 *** First stage dark small round holes (where metal prongs goes in)

228

1

 **** But many false positives, filtered by RF

229

1

 **** Look for pairs of holes

230

1

 *** Second stage does the neighborhood of the found holes look like an outlet

231

1

 ** Need more American data and can train on that

232

1

 * Keep going -- use this as one method of attack.

233

1

234

1

 * 3D box relative to chessboard.  The RCG (*Alexi* (tracking, doorknob) *Maria* (RF, ET))

235

1

 ** Today he's training adaboost on doorknobs

236

1

 ** Error control for planar object position

237

1

 *** Calibration

238

1

239

1

*_Action Items_*

240

1

241

1

*Victor*

242

1

243

1

 * Send images of current outlet detector

244

1

 * Train on Willow images to be sent

245

1

246

1

*Gary*

247

1

248

1

 * Send images of outlets to *Victor* (today)

249

1

 * Try to get several hundred

250

1

 * Get doorknob imagery to *Victor*

251

1

 * Take 3D model of doorknob

252

1

 * CC *Vadim* with Ara

253

1

 * CC to Jean-Bouguet about 4 channel

254

1

 * Talk with Michael about what needs to be done to get C-descriptors in OpenCV

255

1

 * Michael Black's intern

256

1

 * Talk with *Kurt* about moving VSLAM stitching

257

1

258

1

*Vadim*

259

1

260

1

 * Spin images

261

1

 ** Preliminary study is quite fast, could be optmized more

262

1

 ** Most of the time in spin image is not in octree, but in the spin image

263

1

 * Send me OpenCV goals spreadsheet

264

1

 * Put up initial next release goals on OpenCV wiki

265

1

266

1

h2.  2009-01-27

267

1

268

1

*_Agenda_*

269

1

270

1

* Updates

271

1

272

1

*_Minutes_*

273

1

274

1

*_For OpenCV Release_*

275

1

276

1

 * Company Retreat

277

1

 * Outlets

278

1

 ** Bottlenecks

279

1

 *** Find outlets to sub-mm when robot is "in position"

280

1

 **** Must detect that nothing is plugged into the outlet

281

1

 *** Visual servoing after robot touches the wall with the plug

282

1

 *** Finding the plug orientation

283

1

 **** Finding a dropped plug(?)

284

1

 ** Other things which are "good" but not critical for milestone 2

285

1

 *** Detecting Willow Garage outlets from a distance with and without stereo and/or laser data

286

1

 *** Detecting outlets in general from a distance with and without stereo or laser data

287

1

 * Method for first bottleneck (find plugs to sub-mm accuracy)

288

1

 # Using a calibrated camera with rectified images:

289

1

 ## Segment the outlet "ovals", find 4 centroids

290

1

 ## Compute homography to canonical view and remap

291

1

 ## Match templates to find holes and more exact center of outlets

292

1

 ## Re-project more exact center found above into the image to find floating point centers

293

1

 ## Use these reprojected floating point centers to again compute a homography to the canonical view and remap

294

1

 ## Use templates to find the exact location of outlet holes.  This should now be sub-mm precision

295

1

 ## Using the homography and measured outlet, compute the X,Y,Z location of the outlet holes

296

1

 * Arrange for common repository

297

1

298

1

*Vadim*

299

1

300

1

 * Lapack debugged -- now using clapack

301

1

 ** Simple small matrix operations still use native OpenCV code

302

1

 ** Try fast lapack algorithms like ATLAS

303

1

 ** Make use of MKL if available

304

1

 * New memory allocator -- fast for small matrixies  A*B+C

305

1

 *** ~5x faster than malloc and free

306

1

 * Spin images

307

1

 ** 20x speed up on spin images

308

1

 ** Compare against Point Feature Histogram PFH

309

1

 ** Ready in 2-3 more days

310

1

 ** Can be further optimized by factor of 2x ~ 1 week

311

1

 ** 20 seconds on 200K points scene of a conference room using 40cm radius

312

1

 * Stereo

313

1

 ** Textured areas have problems, either use textured light or something like Kolmogorov

314

1

 ** Stereo SSE-2 optimization uses only gcc 4.2 or higher

315

1

 ** Chessboard detector is breaking on high angle views

316

1

 *** High angles are not so useful for calibration, but good for plane finding

317

1

 **** Moderate priority

318

1

 ** Optimizing Kolmogorov

319

1

 *** Sparse graph cuts -- might not work, but idea is to soft fix points that are unlikely to change

320

1

 *** Run as background project for now

321

1

 * Next week

322

1

 ** C-descriptor

323

1

 ** C++ image classes

324

1

 ** cvBoostPatch() from Steve Gould and from Ian Goodfellow

325

1

326

1

*_C descriptor_*

327

1

328

1

In _.../ros/ros-pkg/visino/calonder_descriptor_  (it should be named _C-descriptor_ for "Classifier Descriptor")

329

1

I spoke with Michael Calonder about getting his descriptor (which we use in visual odometry and place recognition) into OpenCV.  He's going to write up a short description of how it works.  It is already fairly optimized using 16 bit vectors and SSE -- it can do a 512x512 image in about 5ms (depending on complexity of course).

330

1

331

1

They've been using the "FAST" detector which has a GPL license.  We might consider doing our own FAST detector.  But, it can use any intrest point detector, it's mainly a really good descriptor.  At the end of training he used Matlab to compress the vectors using PCA -- we'd have to replace that step with clapack.

332

1

333

1

*_Place Recognition/Nistor/Kmeans_*

334

1

335

1

In _../ros/ros-pkg/vision/place_recognition_ there are several things of interest:

336

1

337

1

 # Improved kmeans code (src/kmeans.cpp)

338

1

 # Nistor vocabulary tree which we can add to MLL (src/vocabulary_tree.cpp)

339

1

 # Place recognition  (src/place_recognizer.cpp).  This could be sample code or a really useful app that uses (is an example of use of) the C-descriptor above, along with the vocabulary tree and kmeans.

340

1

341

1

*_Action Items_*

342

1

343

1

*Gary*

344

1

345

1

 * Determine time for outlet meetings with Patrick

346

1

 * Data with plugs already in outlet

347

1

 * Get data to *Victor* of plug images

348

1

 * Put *Victor* in touch with Michael Calonder

349

1

 * Put *Vadim* in touch with Michael about C-descriptor

350

1

 * Find out from *Kurt* status or priority of stereo speedups

351

1

 * Talk to Scott about getting *Vadim* SSH access to machines in Willow

352

1

 * Sandbox in ROS for things like outlet detection experimentation.

353

1

354

1

*Vadim*

355

1

356

1

 * Stereo OpenCV vs *Kurt* algorithm comparison for speed on a known good machine.

357

1

358

1

h2. 2009-02-03

359

1

360

1

*_Agenda_*

361

1

362

1

Outlets will be moved to a meeting directly after the OpenCV meeting, so we will no longer discuss outlet specific issues here but instead focus on OpenCV issues.

363

1

364

1

 * Updates

365

1

 ** *Vadim*

366

1

 ** *Victor*

367

1

 ** *Gary*

368

1

 * Progress towards Summer release

369

1

 ** "OpenCV200906":http://opencv.willowgarage.com/wiki/OpenCV200906

370

1

 ** Memory limitations in machine learning -- design matrix limits

371

1

 ** Error handling???

372

1

 * PnP problem approaches

373

1

374

1

*_Minutes_*

375

1

376

1

*Vadim*

377

1

378

1

# *Anatoly* and me: finished optimization of OctTree & Spinimage code. OpenMP and SSE2 gave ~2x speedup each. In result, the running time of one of our benchmarks dropped from  (ROS was 80 secs), new c code is 880ms and now downto ~230ms with SSE and OpenMP. That is, together with ~20x speedup we achieved with new C code, the total speedup is over 75x. The code has been prepared for integration to OpenCV.

379

1

# The 2 optimizations of the boosting code, proposed by Stephen Gould and Ian Goodfellow, have been integrated to MLL. No feedback from the guys yet.

380

1

# Studied the C-descriptor source code the relevant papers by Michael Calonder and Vincent Lepetit. The algorithm is pretty clear now and the preparation to integration into OpenCV has been started.

381

1

# ~30% of the new matrix class integration has been done. This appeared to be significantly bigger task than I expected.

382

1

# Received the adaptive skin detector code from Farhad Dadgostar; reviewed the code and recieved the corrected variant; The code will be integrated this week.

383

1

384

1

*Victor*

385

1

386

1

 * Outlet detection: calibration, template matching and hole detection have been combined in a single application. The method has been adjusted for another camera (narrow lens resulting in higher resolution outlets). A method for finding a tuple of 4 outlets has been improved for the case of large viewing angle. Recalculation to 3D coordinates in cm scale has been implemented and tested. If an outlet is found correctly the error of hole detection is less than 0.5mm for all cases but one.

387

1

 * Extreme trees (*Maria*): the code has been refactored, all ET functionality has been encapsulated into a separate class inherited from CvRTrees. The performance of RF and ET has been compared. For very deep trees (depth 20) the performance is comparable, while for more shallow trees (less than 10) ET is faster at least 2x.

388

1

 ** Has natural parallelism, look at it.

389

1

 ** Parallize all boosted and random prediction trees.

390

1

 ** Possible paper in statistical convergence to correct answer

391

1

 * 3D object tracking (*Alex*ey): the method for tracking 3D will be based on 5 points set by user on two different frames. Stereo pair calibration has been implemented.

392

1

 ** Right now we are trying to catch a bug in calculating 3D from disparity.  *Alexi* out so this will wait.

393

1

394

1

Other: *Alex*ey is on vacation Feb2-8, back in the office Feb 9.

395

1

396

1

*Gary*

397

1

398

1

Boosting

399

1

400

1

 * Will try to see if we're still artificially limited by memory size

401

1

402

1

OpenCV errors

403

1

404

1

 * New implmentation moves completely to exceptions

405

1

 * Should work with debugger now

406

1

407

1

OpenCV website

408

1

409

1

 * For demos etc

410

1

411

1

*_Action Items_*

412

1

413

1

*_NEW_*

414

1

415

1

*Gary*

416

1

417

1

 * -Email Steve Gould about boosting memory limits with *Victor* and *Vadim*-

418

1

 * -Email Ian Goodfellow about opencv error catching with *Vadim*.-

419

1

420

1

*Vadim*

421

1

422

1

 * Check ML design matrix

423

1

424

1

*Victor*

425

1

426

1

*_FROM LAST TIME_*

427

1

428

1

*Gary*

429

1

430

1

 * Determine time for outlet meetings with Patrick

431

1

 ** 10am (9pm NN) on Tuesday

432

1

 * Data with plugs already in outlet

433

1

 * Get data to *Victor* of plug images

434

1

 * Put *Victor* in touch with Michael Calonder

435

1

 * Put *Vadim* in touch with Michael about C-descriptor

436

1

 * (?) Find out from *Kurt* status or priority of stereo speedups

437

1

 * (?) Talk to Scott about getting *Vadim* SSH access to machines in Willow

438

1

 ** I talked, but no action yet.

439

1

 * Sandbox in ROS for things like outlet detection experimentation.

440

1

 ** Just  use sourceforge and ROS can pull it in if you build say the outlet node.

441

1

442

1

*Vadim*

443

1

444

1

 * Stereo OpenCV vs *Kurt* algorithm comparison for speed on a known good machine.

445

1

446

1

h2. 2009-02-10

447

1

448

1

*_Agenda_*

449

1

450

1

Quick updates:

451

1

452

1

*Gary*

453

1

454

1

 * Helped get "Mech Turk interface":http://pr.willowgarage.com/wiki/ROS/mturk into Willow Garage (thanks to *Alexander* Sorokin and Scott Hassan's 2 daze of hacking last week). View some "results":http://vm6.willowgarage.com:8080/mt/results/ (click on the numbers under "completed hits").

455

1

 ** We can use this for massive data sets and even say sponsor people to create better face recognizers etc.

456

1

457

1

*Vadim*

458

1

459

1

 * Matrix interface

460

1

 * 80% of cxcore is ported to the new C++ interface

461

1

 ** Automatic garbage collection (reference counter based)

462

1

 ** Can use with STL vector of images and matrixes

463

1

 ** Have template interface matrix(i,j) does the right thing

464

1

 *** Thread safety coming, can use atomic operations, efficient on Intel and AMD archs

465

1

 * IPP will now be installed/built at configure instead of dynamic

466

1

 ** Using cmake, if IPP is detected the build flag will be on by default, user can explicitly turn it off (same way as ffmpg libs etc)

467

1

 * Decision tree memory optimization, needs about 2 weeks (*Maria* knows the code a bit, but *Vadim* expert)

468

1

 ** Needed to move cascade recognizer to general decision trees, beyond Haar features

469

1

 ** Think about collaborating with someone to train good working models for the release

470

1

 * Spin images are done but waiting to complete the C++ integration before integration (probably by Friday)

471

1

 * Focus results, ... lower priority

472

1

473

1

*Victor*

474

1

475

1

 * Extreme Trees

476

1

 ** Categorical support is added

477

1

 ** Don't need sorting of variables, so lower memory requirements

478

1

 ** Can handle 100K samples of 1000 features, Extreme trees worked OK

479

1

 ** Code is in SVN

480

1

 ** Recognition performance is comparable to RF but faster

481

1

 *** Inherently more parallel

482

1

 *** Try on outlets?

483

1

 * 3D object tracking (*Alexander* was on vacation)

484

1

 ** Estimate pose of box in space from motion and knowlege of box

485

1

 ** Function for disparity to 3D has bug

486

1

 ** 2D version ... finished ... but special case of 3D, integrate

487

1

488

1

489

1

OpenCV release tracking

490

1

491

1

 * Go over the "development plans":http://opencv.willowgarage.com/wiki/OpenCV200906

492

1

493

1

*_Minutes_*

494

1

495

1

Tracking the summer release:

496

1

497

1

 # General Structure:

498

1

 ** Decisions on library structure (Google's request, separate packages as is ...)

499

1

 *** Provide way to build OpenCV statically, then size varies with binary size

500

1

 # New Features

501

1

 ** C++ interface

502

1

 *** ... see above

503

1

 ** Better python

504

1

 *** *Vadim*, *Alex* ...

505

1

 *** C++ will take another month but can improve python in parallel

506

1

 ** 2D features C-descriptor, HoG Boundary Fragments ...

507

1

 *** C-descriptor ... resolved issues with Michael, *Vadim* understands code more deeply (next week)

508

1

 *** Boundary fragments ... working on focus.

509

1

 *** Get contributions of 2D features externally

510

1

 *** MSER ... (*Gary* has code)

511

1

 *** Star is already in and in for ROS build

512

1

 ** Alpha channel, alternatives to contours RLE

513

1

 *** Introduce functions with advance ways of blending

514

1

 *** Mask channels

515

1

 *** Optical flow

516

1

 ** 3D features (Plane fit, PFH, FPFH from Radu), 3D Delaunay, 3D convex hull, mesh generation

517

1

 *** SPIN almost done

518

1

 *** *Radu* is organizing his code into a library ... maybe we can just lift this

519

1

 ** VSLAM

520

1

 *** Wait on what *Kurt* and *James* get together

521

1

 ** Image stitching

522

1

 *** will ask Jean-Yves for pointers

523

1

 ** Optical flow

524

1

 *** Move code derived form former work in OpenCV

525

1

 ** MLL

526

1

 *** Extreme trees are coming

527

1

 *** Reduce memory footprint

528

1

 *** Need to work on input and output of data in CSV (or other simple deliminators), XML and YML

529

1

 *** Regression tests ... maybe general cross validation

530

1

 ** Misc: Focus detector, adaptive skin tracking

531

1

 *** As discussed above

532

1

 # Optimizations

533

1

 ** .

534

1

 # Documenation

535

1

 ** .

536

1

 # High GUI

537

1

 ** .

538

1

539

1

*_Action Items_*

540

1

541

1

*_NEW_*

542

1

543

1

Meeting next week will be on Wednesday

544

1

545

1

*Gary*

546

1

547

1

 * Send *Vadim* MSER code

548

1

 * Ask Jean-Yves for priority list of functions to work with mask and alpha channels

549

1

 * Ask Jean-Yves about image stitching support/libraries/functionality

550

1

 * Ask *Radu* about his library progress.

551

1

 * Send email reminder for Wednesday meeting next week.

552

1

553

1

*Vadim*

554

1

555

1

 * Adaptive skin tracker app/test

556

1

 * Finish cxcore rewrite, make sure all the tests pass, update the documentation.

557

1

 * Start cv module revision.

558

1

 * Integrate spin images.

559

1

560

1

*Victor*

561

1

562

1

*_FROM LAST TIME_*

563

1

564

1

*Gary*

565

1

566

1

 * -Email Steve Gould about boosting memory limits with *Victor* and *Vadim*-

567

1

 * -Email Ian Goodfellow about opencv error catching with *Vadim*.-

568

1

569

1

*Vadim*

570

1

571

1

 * Check ML design matrix

572

1

573

1

*Victor*

574

1

575

1

h2. 2009-02-18

576

1

577

1

*_Agenda_*

578

1

579

1

Accomplishment Tracking

580

1

581

1

* .

582

1

583

1

Chessboard issues

584

1

585

1

 * Detection failures

586

1

 * 2 functions? One chessboards for calibration, another chessboards for pose?

587

1

588

1

3D capture, discuss implementation of

589

1

590

1

 * http://www.robots.ox.ac.uk/~awf/turntable/index.html

591

1

592

1

cxcore re-write, cv-rewrite

593

1

594

1

Spin, skin

595

1

596

1

Release tracking.

597

1

598

1

*_Minutes_*

599

1

600

1

_*Tracking the summer release:*_

601

1

602

1

 # General Structure:

603

1

 ** Decisions on library structure (Google's request, separate packages as is ...)

604

1

 *** Provide way to build OpenCV statically, then size varies with binary size

605

1

 # New Features

606

1

 ** C++ interface

607

1

 *** ... see below

608

1

 ** Better python

609

1

 *** *Vadim*, *Alex* ...

610

1

 *** C++ will take another month but can improve python in parallel

611

1

 ** 2D features C-descriptor, HoG Boundary Fragments ...

612

1

 *** C-descriptor ... resolved issues with Michael, *Vadim* understands code more deeply (next week)

613

1

 *** Boundary fragments ... working on focus.

614

1

 *** Get contributions of 2D features externally

615

1

 *** MSER ... (*Gary* has code)

616

1

 *** Star is already in and in for ROS build

617

1

 ** Alpha channel, alternatives to contours RLE

618

1

 *** Introduce functions with advance ways of blending

619

1

 *** Mask channels

620

1

 *** Optical flow

621

1

 ** 3D features (Plane fit, PFH, FPFH from Radu), 3D Delaunay, 3D convex hull, mesh generation

622

1

 *** SPIN almost done

623

1

 *** *Radu* is organizing his code into a library ... maybe we can just lift this

624

1

 ** VSLAM

625

1

 *** Wait on what *Kurt* and *James* get together

626

1

 ** Image stitching

627

1

 *** will ask Jean-Yves for pointers

628

1

 ** Optical flow

629

1

 *** Move code derived form former work in OpenCV

630

1

 ** MLL

631

1

 *** Extreme trees are coming

632

1

 *** Reduce memory footprint

633

1

 *** Need to work on input and output of data in CSV (or other simple deliminators), XML and YML

634

1

 *** Regression tests ... maybe general cross validation

635

1

 ** Misc: Focus detector, adaptive skin tracking

636

1

 *** As discussed above

637

1

 # Optimizations

638

1

 ** .

639

1

 # Documentation

640

1

 ** .

641

1

 # High GUI

642

1

 ** .

643

1

644

1

_*New Interface*_

645

1

646

1

 * cxcore revision is complete (except for XML/YAML input/output functions). All the algorithmic tests for cxcore and cv pass, all the samples run well.

647

1

 ** As a small illustration of the new API capabilities, here is a some code that I used to debug SVD:

648

1

649

1

 <pre><code class="cpp">

650

1

    // matrix A comes in

651

1

    SVD svd(A);

652

1

    int n = A.rows;

653

1

654

1

    double a = norm(svd.u*svd.u.t() - Mat::eye(n,n), CV_C);

655

1

    double b = norm(svd.vt*svd.vt.t() - Mat::eye(n,n), CV_C);

656

1

    int nz = countNonZero(svd.w >= 0);

657

1

    double c = norm(svd.u*Mat::diag(svd.w)*svd.vt - Mat::eye(n,n), CV_C);

658

1

    CV_Assert( a < 1e-5 && b < 1e-5 && c < 1e-5 && nz == n );

659

1

</code></pre>

660

1

661

1

 * Can template this

662

1

 ** Can create image with up to 64 channels

663

1

 ** Some functions limit to 4 channels

664

1

 ** Splitting channels is easy

665

1

 *** Work with IPP, for cache, want to extract channel

666

1

667

1

_*General Issues:*_

668

1

669

1

Chessboard

670

1

671

1

 * cvFindCornersSubpix  fails with large search window, diverges.

672

1

 ** Ordering of chessboard corners and splitting

673

1

 ** Will lookat

674

1

 * Initial location of corners can be wrong

675

1

 ** Extreme angles can create problems for morphological operations

676

1

 *** vertical and horizontal operations can make distance between corners smaller than within squares

677

1

678

1

3D

679

1

680

1

 * http://www.robots.ox.ac.uk/~awf/turntable/index.html

681

1

Skin detection should work using highGUI camera

682

1

683

1

Spin images, trying on Ikea

684

1

685

1

3D tracking (*Victor*, *Alexi*) (3D box)

686

1

687

1

 * 3d positions of markers using Chessboard

688

1

 ** Currently unstable depending on where the user clicks

689

1

 ** Somewhere in disparity to 3D transformation

690

1

 ** Generating tests on synthetic data by tomorrow.

691

1

692

1

Memory use with MLL

693

1

694

1

 * Shrank memory use several times

695

1

 * Debugging

696

1

697

1

*_Action Items_*

698

1

699

1

*_NEW_*

700

1

701

1

ALL

702

1

703

1

 * Send in list of what's been accomplished so far in OpenCV since Willow started supporting development.

704

1

705

1

*Gary*

706

1

707

1

 * Look into sending prosilica and other cameras to Argus

708

1

 * Send full data link with mask images for Spin image work

709

1

 * Get Zisserman 3D paper to *Vadim* (on ACM)

710

1

711

1

*Vadim*

712

1

713

1

 * Look at better calibration and pose object

714

1

 ** Equal or better in accuracy and stability

715

1

 * Check what you think 3D scanning would take

716

1

717

1

*Victor*

718

1

719

1

*_FROM LAST TIME_*

720

1

721

1

Meeting next week will be on Wednesday

722

1

723

1

*Gary*

724

1

725

1

 * Send *Vadim* MSER code

726

1

 * [but he didn't answer yet] Ask Jean-Yves for priority list of functions to work with mask and alpha channels

727

1

 * [same] Ask Jean-Yves about image stitching support/libraries/functionality

728

1

 * Ask *Radu* about his library progress. Radu:

729

1

<pre>

730

1

We took a few steps today to unify the point cloud -related libraries in ROS under one umbrella. The result is that

731

1

there is now a _point_cloud_mapping_ package which contains the following things:

732

1

- cloud_geometry: areas, distances, intersections, 3D normals/features estimation, norms, projections, statistics,

733

1

transformations, etc;

734

1

- cloud_io: a library for Input/Output PCD (Point Cloud Data) processing;

735

1

- cloud_kdtree: a library for nD spatial decomposition and nearest neighbors searches for PointCloud messages using k-D

736

1

trees;

737

1

- cloud_octree: a library for 3D spatial decomposition of PointCloud messages using octree data structures;

738

1

- sample_consensus: a library for SAmple Consensus (SAC) like methods (RANSAC, MSAC, MLESAC, etc,) and geometric shape

739

1

models;

740

1

There are also 2 basic nodes included with the package:

741

1

- cloud_downsampler_node: a node which uniformly downsamples a 3D point cloud

742

1

- normal_estimation_node: a node which estimates local surface properties at each point such as surface normals,

743

1

curvature changes, moment invariants, etc.

744

1

</pre>

745

1

 * -Send email reminder for Wednesday meeting next week.-

746

1

747

1

 *Vadim*

748

1

749

1

 * Adaptive skin tracker app/test

750

1

 * Finish cxcore rewrite, make sure all the tests pass, update the documentation.

751

1

 * Start cv module revision.

752

1

 * Integrate spin images.

753

1

754

1

*Victor*

755

1

756

1

h2. 2009-02-24

757

1

758

1

*_Agenda_*

759

1

760

1

 * Accomplishments

761

1

 * Interface update

762

1

 * Chessboard improvements

763

1

 * 3D Box relative to chessboard

764

1

 * 3D Object Capture paper/analysis

765

1

 * Spin images

766

1

 * Summer release tracking

767

1

768

1

*_Minutes_*

769

1

770

1

_*Tracking the summer release:*_

771

1

772

1

 # General Structure:

773

1

 ** Decisions on library structure (Google's request, separate packages as is ...)

774

1

 *** Provide way to build OpenCV statically, then size varies with binary size

775

1

 # New Features

776

1

 ** C++ interface

777

1

 *** cxcore

778

1

 *** cv  -- start with filtering functions, resize, optical flow, object detection (2 weeks)

779

1

 ** Better python

780

1

 *** *Vadim*, *Alex* ...

781

1

 *** C++ will take another month but can improve python in parallel

782

1

 ** 2D features C-descriptor, HoG Boundary Fragments ...

783

1

 *** C-descriptor ... resolved issues with Michael, *Vadim* understands code more deeply (next week)

784

1

 **** No re-implementation yet. Code based on randomized tree, but specialized version.

785

1

 *** Boundary fragments ... working on focus.

786

1

 *** Get contributions of 2D features externally

787

1

 *** MSER ... (*Gary* has code)

788

1

 *** HoG

789

1

 *** Star is already in and in for ROS build

790

1

 ** Alpha channel, alternatives to contours RLE

791

1

 *** Introduce functions with advance ways of blending

792

1

 *** Mask channels

793

1

 *** Optical flow

794

1

 ** 3D features (Plane fit, PFH, FPFH from Radu), 3D Delaunay, 3D convex hull, mesh generation

795

1

 *** SPIN almost done

796

1

 *** *Radu* is organizing his code into a library ... maybe we can just lift this

797

1

 ** VSLAM

798

1

 *** Wait on what *Kurt* and *James* get together

799

1

 ** Image stitching

800

1

 *** will ask Jean-Yves for pointers

801

1

 ** Optical flow

802

1

 *** Move code derived form former work in OpenCV

803

1

 ** MLL

804

1

 *** Extreme trees are coming

805

1

 *** Reduce memory footprint

806

1

 *** Need to work on input and output of data in CSV (or other simple deliminators), XML and YML

807

1

 *** Regression tests ... maybe general cross validation

808

1

 *** Generalize cascade

809

1

 ** Misc: Focus detector, adaptive skin tracking

810

1

 *** As discussed above

811

1

 # Optimizations

812

1

 ** .

813

1

 # Documentation

814

1

 ** .

815

1

 # High GUI

816

1

 ** .

817

1

818

1

_*OTHER*_

819

1

820

1

New Interface

821

1

822

1

 * Whole new interface for cxcore is finsished

823

1

 ** Moving to cv

824

1

 ** Examples: Update samples.  .c code will be old interface, .cpp will be new interface

825

1

 ** Output is simplified by automatically correct type

826

1

 ** Start effort to doxygen (lower background priority)

827

1

828

1

Chessboard code that

829

1

830

1

 * *Vadim* has released a patch that seems to fix all known problems.  Congrats!

831

1

 * High angle (60 degrees) boards still cannot be done

832

1

 * Detect lines, automatically detect board size, perhaps remap image to frontal view.

833

1

 ** Sub-pixel finding is now put inside the algorithm rather than after.

834

1

835

1

3D box

836

1

837

1

 * 3D object tracking: the code is in debugging phase. The calibration of a stereo pair is still unstable, producing large reprojection errors too often.

838

1

 * *Alexi* not much time to work on in

839

1

 * Make relative to chessboard

840

1

841

1

842

1

3D Object Capture

843

1

844

1

 * Bundle adjustment missing from OpenCV

845

1

 * Mesh reconstruction, not in OpenCV

846

1

 * Might take a month if we can't get the above code

847

1

 ** http://www.3dsom.com/index.html  seems to use the same technique.

848

1

849

1

Spin images

850

1

851

1

 * Not incorporated yet, code is ready, just time to integrate

852

1

 ** Test code is ready

853

1

 ** 1 or 2 days

854

1

855

1

Extremely random trees

856

1

857

1

 * Put a callback in the splitting function

858

1

859

1

MSER

860

1

861

1

 * *Alex* has MSER code

862

1

863

1

*_Action Items_*

864

1

865

1

*_NEW_*

866

1

867

1

*Gary*

868

1

869

1

 * Ask *Kurt* about where Bundle adjustment is ... if it is

870

1

 * -Send *Victor* Florian's paper-

871

1

872

1

*Vadim*

873

1

874

1

 * Doxygen documentation as background, lower priority task.

875

1

876

1

*Victor*

877

1

878

1

*_FROM LAST TIME_*

879

1

880

1

ALL

881

1

882

1

 * -Send in list of what's been accomplished so far in OpenCV since Willow started supporting development.-

883

1

884

1

*Gary*

885

1

886

1

 * Look into sending prosilica and other cameras to Argus

887

1

 * -Send full data link with mask images for Spin image work-

888

1

 * -Get Zisserman 3D paper to *Vadim* (on ACM)-

889

1

890

1

*Alexi*

891

1

892

1

 * Finish up 3D box

893

1

894

1

*Vadim*

895

1

896

1

 * Look at better calibration and pose object

897

1

 ** Equal or better in accuracy and stability

898

1

 * Check what you think 3D scanning would take

899

1

900

1

*Victor*

901

1

902

1

h2. 2009-02-24

903

1

904

1

*_Agenda_*

905

1

906

1

 * Accomplishments

907

1

 * Interface update

908

1

 * Chessboard improvements

909

1

 * 3D Box relative to chessboard

910

1

 * 3D Object Capture paper/analysis

911

1

 * Spin images

912

1

 * Summer release tracking

913

1

914

1

*_Minutes_*

915

1

916

1

_*Tracking the summer release:*_

917

1

918

1

 # General Structure:

919

1

 ** Decisions on library structure (Google's request, separate packages as is ...)

920

1

 *** Provide way to build OpenCV statically, then size varies with binary size

921

1

 # New Features

922

1

 ** C++ interface

923

1

 *** cxcore

924

1

 *** cv  -- start with filtering functions, resize, optical flow, object detection (2 weeks)

925

1

 ** Better python

926

1

 *** *Vadim*, *Alex* ...

927

1

 *** C++ will take another month but can improve python in parallel

928

1

 ** 2D features C-descriptor, HoG Boundary Fragments ...

929

1

 *** C-descriptor ... resolved issues with Michael, *Vadim* understands code more deeply (next week)

930

1

 **** No re-implementation yet. Code based on randomized tree, but specialized version.

931

1

 *** Boundary fragments ... working on focus.

932

1

 *** Get contributions of 2D features externally

933

1

 *** MSER ... (*Gary* has code)

934

1

 *** HoG

935

1

 *** Star is already in and in for ROS build

936

1

 ** Alpha channel, alternatives to contours RLE

937

1

 *** Introduce functions with advance ways of blending

938

1

 *** Mask channels

939

1

 *** Optical flow

940

1

 ** 3D features (Plane fit, PFH, FPFH from Radu), 3D Delaunay, 3D convex hull, mesh generation

941

1

 *** SPIN almost done

942

1

 *** *Radu* is organizing his code into a library ... maybe we can just lift this

943

1

 ** VSLAM

944

1

 *** Wait on what *Kurt* and *James* get together

945

1

 ** Image stitching

946

1

 *** will ask Jean-Yves for pointers

947

1

 ** Optical flow

948

1

 *** Move code derived form former work in OpenCV

949

1

 ** MLL

950

1

 *** Extreme trees are coming

951

1

 *** Reduce memory footprint

952

1

 *** Need to work on input and output of data in CSV (or other simple deliminators), XML and YML

953

1

 *** Regression tests ... maybe general cross validation

954

1

 *** Generalize cascade

955

1

 ** Misc: Focus detector, adaptive skin tracking

956

1

 *** As discussed above

957

1

 # Optimizations

958

1

 ** .

959

1

 # Documentation

960

1

 ** .

961

1

 # High GUI

962

1

 ** .

963

1

964

1

_*OTHER*_

965

1

966

1

New Interface

967

1

968

1

 * Whole new interface for cxcore is finsished

969

1

 ** Moving to cv

970

1

 ** Examples: Update samples.  .c code will be old interface, .cpp will be new interface

971

1

 ** Output is simplified by automatically correct type

972

1

 ** Start effort to doxygen (lower background priority)

973

1

974

1

Chessboard code that

975

1

976

1

 * *Vadim* has released a patch that seems to fix all known problems.  Congrats!

977

1

 * High angle (60 degrees) boards still cannot be done

978

1

 * Detect lines, automatically detect board size, perhaps remap image to frontal view.

979

1

 ** Sub-pixel finding is now put inside the algorithm rather than after.

980

1

981

1

3D box

982

1

983

1

 * 3D object tracking: the code is in debugging phase. The calibration of a stereo pair is still unstable, producing large reprojection errors too often.

984

1

 * *Alexi* not much time to work on in

985

1

 * Make relative to chessboard

986

1

987

1

988

1

3D Object Capture

989

1

990

1

 * Bundle adjustment missing from OpenCV

991

1

 * Mesh reconstruction, not in OpenCV

992

1

 * Might take a month if we can't get the above code

993

1

 ** http://www.3dsom.com/index.html  seems to use the same technique.

994

1

995

1

Spin images

996

1

997

1

 * Not incorporated yet, code is ready, just time to integrate

998

1

 ** Test code is ready

999

1

 ** 1 or 2 days

1000

1

1001

1

Extremely random trees

1002

1

1003

1

 * Put a callback in the splitting function

1004

1

1005

1

MSER

1006

1

1007

1

 * *Alex* has MSER code

1008

1

1009

1

*_Action Items_*

1010

1

1011

1

*_NEW_*

1012

1

1013

1

*Gary*

1014

1

1015

1

 * Ask *Kurt* about where Bundle adjustment is ... if it is

1016

1

 * -Send *Victor* Florian's paper-

1017

1

1018

1

*Vadim*

1019

1

1020

1

 * Doxygen documentation as background, lower priority task.

1021

1

1022

1

*Victor*

1023

1

1024

1

*_FROM LAST TIME_*

1025

1

1026

1

ALL

1027

1

1028

1

 * -Send in list of what's been accomplished so far in OpenCV since Willow started supporting development.-

1029

1

1030

1

*Gary*

1031

1

1032

1

 * Look into sending prosilica and other cameras to Argus

1033

1

 * -Send full data link with mask images for Spin image work-

1034

1

 * -Get Zisserman 3D paper to *Vadim* (on ACM)-

1035

1

1036

1

*Alexi*

1037

1

1038

1

 * Finish up 3D box

1039

1

1040

1

*Vadim*

1041

1

1042

1

 * Look at better calibration and pose object

1043

1

 ** Equal or better in accuracy and stability

1044

1

 * Check what you think 3D scanning would take

1045

1

1046

1

*Victor*

1047

1

1048

1

h2. 2009-03-03

1049

1

1050

1

*_Agenda_*

1051

1

1052

1

 * SURF and *_OpenCV Announce_*

1053

1

 * *Marius* Muja is at Willow for 6 months

1054

1

 ** FLANN incorporation

1055

1

 * New interface update

1056

1

 * 3D Box -- it slips

1057

1

 * 3D Model

1058

1

 * MSER

1059

1

1060

1

*_Minutes_*

1061

1

1062

1

_*Tracking the summer release:*_

1063

1

1064

1

 # General Structure:

1065

1

 ** Decisions on library structure (Google's request, separate packages as is ...)

1066

1

 *** Provide way to build OpenCV statically, then size varies with binary size

1067

1

 # New Features

1068

1

 ** C++ interface

1069

1

 *** cxcore

1070

1

 *** cv  -- start with filtering functions, resize, optical flow, object detection (2 weeks)

1071

1

 ** Better python

1072

1

 *** *Vadim*, *Alex* ...

1073

1

 *** C++ will take another month but can improve python in parallel

1074

1

 ** 2D features C-descriptor, HoG Boundary Fragments ...

1075

1

 *** C-descriptor ... resolved issues with Michael, *Vadim* understands code more deeply (next week)

1076

1

 **** No re-implementation yet. Code based on randomized tree, but specialized version.

1077

1

 *** Boundary fragments ... working on focus.

1078

1

 *** Get contributions of 2D features externally

1079

1

 *** MSER ... (*Gary* has code)

1080

1

 *** HoG

1081

1

 *** Star is already in and in for ROS build

1082

1

 ** Alpha channel, alternatives to contours RLE

1083

1

 *** New alpha transparency (per pixel) added

1084

1

 *** Introduce functions with advance ways of blending

1085

1

 *** Mask channels

1086

1

 *** Optical flow

1087

1

 ** 3D features (Plane fit, PFH, FPFH from Radu), 3D Delaunay, 3D convex hull, mesh generation

1088

1

 *** SPIN almost done

1089

1

 *** *Radu* is organizing his code into a library ... maybe we can just lift this

1090

1

 ** 3D capture  http://www.3dsom.com/index.html

1091

1

 ** VSLAM

1092

1

 *** Wait on what *Kurt* and *James* get together

1093

1

 ** Image stitching

1094

1

 *** will ask Jean-Yves for pointers

1095

1

 ** Optical flow

1096

1

 *** Move code derived form former work in OpenCV

1097

1

 ** MLL

1098

1

 *** Extreme trees are coming

1099

1

 *** Reduce memory footprint

1100

1

 *** Need to work on input and output of data in CSV (or other simple deliminators), XML and YML

1101

1

 *** Regression tests ... maybe general cross validation

1102

1

 *** Generalize cascade

1103

1

 ** Misc: Focus detector, adaptive skin tracking

1104

1

 *** As discussed above

1105

1

 # Optimizations

1106

1

 ** .

1107

1

 # Documentation

1108

1

 ** .

1109

1

 # High GUI

1110

1

 ** .

1111

1

1112

1

_*OTHER*_

1113

1

1114

1

OpenCV Announce ...

1115

1

1116

1

 * SURF is in #1 in change log

1117

1

 * Post change log to wiki

1118

1

 * OpenCV Weekly ...

1119

1

1120

1

FLANN

1121

1

1122

1

 * Some timing results at bottom

1123

1

 * Code is in good shape

1124

1

 ** 3D party directoy can be put there

1125

1

 ** For mac, can be put in system path

1126

1

 ** Linux can select

1127

1

 * In CV module, can create a wrapper for CvMat interface to FLANN

1128

1

 ** Make test, so can use included version of FLANN

1129

1

 ** First looks to internal unless user specifies external

1130

1

 *** Put flag to download external

1131

1

 *** We keep reasonably up to date with external

1132

1

1133

1

Interface update

1134

1

1135

1

 * cxcore in SVN

1136

1

 ** Mac and Windows,

1137

1

 ** Linux: gcc caused changes due to template handling differences

1138

1

 * convolution in cv

1139

1

 ** Then geometrical transforms

1140

1

 ** Optical flow

1141

1

 ** Add new alpha blending function with separate transparency for each pixel

1142

1

 ** Further functions can be lower priority

1143

1

1144

1

3D box

1145

1

1146

1

 * Box slips up (pretty sure)

1147

1

 * Image is rectified

1148

1

 ** If camera is close, distortion can be large

1149

1

 ** Tomorrow will ship new video with box being tracked

1150

1

 * Could be chessboard slippage ?

1151

1

1152

1

3D Object Capture

1153

1

1154

1

 * Bundle adjustment missing from OpenCV

1155

1

 * Mesh reconstruction, not in OpenCV

1156

1

 * Might take a month if we can't get the above code

1157

1

 ** http://www.3dsom.com/index.html  seems to use the same technique.

1158

1

 * Initialization of extrinsics (pose) is hardest part of the code

1159

1

 * Different from general SFM or stereo

1160

1

 * smart inversion of J^t*J

1161

1

 ** pass to SVD (might be 10x slower but easier)

1162

1

1163

1

MSER

1164

1

1165

1

 * *Gary* sent

1166

1

 ** *Victor* looking at right now,

1167

1

 *** *Alex* doing his own version faster at lower priority ~ couple of weeks

1168

1

1169

1

SURF

1170

1

1171

1

 * detection and descriptor united

1172

1

 ** Should be split

1173

1

 ** "Find_obj.cpp" is demo code in samples/c

1174

1

1175

1

*Vadim*:

1176

1

1177

1

 # cxcore has been built with GCC 4.0 and GCC 4.2 on MacOSX. Quite a few changes have been made because GCC handles some template stuff differently and usually catches more bugs. Another portion of changes were made to make all the tests pass. This tested version of cxcore has been committed to SVN repository.

1178

1

 # The work on revising cv has been started, new interface for filtering functions (linear separable and non-separable, Laplacian, median, morphogical operations) is ~90% ready (the code was written and now debugged).

1179

1

 # fixed 1 minor bug in boosting (reported by Ian Goodfellow) and 2 bugs in camera calibration (reported by Sergey).

1180

1

 # *Anatoly* started the work on benchmarking various fast n-dim search algorithms for quality & speed. For now we have 3 algorithms in OpenCV: kdtree, spilltree, LSH (locally-sensitive hash), and there is also FLANN. We could not run LSH properly yet, but here are some first results on synthetic uniformly-distributed data in n-dim cubes:

1181

1

 ** tAvgDist is the average (over the test set) distance to the nearest point, as returned by various alorithms (DumbBruteForce is the naive algorithm), the closer to the naive algorithm, the better.

1182

1

 ** M1 is average over the test set of (min_dist/average_dist), the smaller the better.

1183

1

 ** M2 is the number of times the returned point was closer than MAX(average_dist - stddev(distance), 0)

1184

1

 ** and M3 is average of (average_dist - min_dist)/stddev(distance). More than 3 should good enough from statistical point of view, but the larger - the better.

1185

1

 * These preliminary results show that kd-tree is the fastest method and generally not that bad, FLANN is as accurate as naive algorithm, yet significantly faster (but significantly slower than kd-tree, we wonder, whether it can be tuned to be less accurate but faster)

1186

1

 * spill tree is only good for low-dimensional space, otherwise brute force is much faster.

1187

1

 * LSH is yet to be tested.

1188

1

1189

1

*_Action Items_*

1190

1

1191

1

*_NEW_*

1192

1

1193

1

*Gary*

1194

1

1195

1

 * -Make change log wiki page, and send to *Vadim*-

1196

1

 * -Make OpenCV Montly Newsletter and send to *Vadim*-

1197

1

 * Ask J-Y again about mask image uses/image stitching

1198

1

 * -Find JD's code and send to *Vadim*-

1199

1

 * -Send email to *Radu* to point to mesh code/techniques-

1200

1

 * Calendor to Andrew Ng

1201

1

1202

1

*Vadim*

1203

1

1204

1

 * Put up change log wiki page and monthly page after *Gary* set's up

1205

1

1206

1

*Victor*

1207

1

1208

1

 * Do new video of box

1209

1

1210

1

*_FROM LAST TIME_*

1211

1

1212

1

*Gary*

1213

1

1214

1

 * Ask *Kurt* about where Bundle adjustment is ... if it is

1215

1

 ** _Umm, JD Chen (no longer with Willow) supposedly did this, but for stereo code, sent him an email_

1216

1

 * -Send *Victor* Florian's paper-

1217

1

 ** http://www.robots.ox.ac.uk/~vgg/publications/papers/schroff08.pdf

1218

1

 ** http://www.robots.ox.ac.uk/~vgg/publications/papers/schroff06.pdf

1219

1

1220

1

*Vadim*

1221

1

1222

1

 * Doxygen documentation as background, lower priority task.

1223

1

1224

1

*Victor*

1225

1

1226

1

h2. 2009-03-10

1227

1

1228

1

*_Agenda_*

1229

1

1230

1

Summer of Code

1231

1

1232

1

New interface status

1233

1

1234

1

FLANN Benchmarks

1235

1

1236

1

Status of Extremely Random Trees -- is it in

1237

1

1238

1

MSER?

1239

1

1240

1

3D Box

1241

1

1242

1

3D capture

1243

1

 * Bundle adjustment missing from OpenCV

1244

1

 * Mesh reconstruction, not in OpenCV

1245

1

 * Might take a month if we can't get the above code

1246

1

 ** http://www.3dsom.com/index.html  seems to use the same technique.

1247

1

1248

1

*_Minutes_*

1249

1

1250

1

_*Tracking the summer release:*_

1251

1

1252

1

 # General Structure:

1253

1

 ** Decisions on library structure (Google's request, separate packages as is ...)

1254

1

 *** Provide way to build OpenCV statically, then size varies with binary size

1255

1

 # New Features

1256

1

 ** C++ interface

1257

1

 *** cxcore

1258

1

 *** cv  -- start with filtering functions, resize, optical flow, object detection (2 weeks)

1259

1

 ** Better python

1260

1

 *** *Vadim*, *Alex* ...

1261

1

 *** C++ will take another month but can improve python in parallel

1262

1

 ** 2D features C-descriptor, HoG Boundary Fragments ...

1263

1

 *** C-descriptor ... resolved issues with Michael, *Vadim* understands code more deeply (next week)

1264

1

 **** No re-implementation yet. Code based on randomized tree, but specialized version.

1265

1

 *** Boundary fragments ... working on focus.

1266

1

 *** Get contributions of 2D features externally

1267

1

 *** MSER ... (*Gary* has code)

1268

1

 *** HoG

1269

1

 *** Star is already in and in for ROS build

1270

1

 ** Alpha channel, alternatives to contours RLE

1271

1

 *** New alpha transparency (per pixel) added

1272

1

 *** Introduce functions with advance ways of blending

1273

1

 *** Mask channels

1274

1

 *** Optical flow

1275

1

 ** 3D features (Plane fit, PFH, FPFH from Radu), 3D Delaunay, 3D convex hull, mesh generation

1276

1

 *** SPIN almost done

1277

1

 *** *Radu* is organizing his code into a library ... maybe we can just lift this

1278

1

 ** 3D capture  http://www.3dsom.com/index.html

1279

1

 ** VSLAM

1280

1

 *** Wait on what *Kurt* and *James* get together

1281

1

 ** Image stitching

1282

1

 *** will ask Jean-Yves for pointers

1283

1

 ** Optical flow

1284

1

 *** Move code derived form former work in OpenCV

1285

1

 ** MLL

1286

1

 *** Extreme trees are coming

1287

1

 *** Reduce memory footprint

1288

1

 *** Need to work on input and output of data in CSV (or other simple deliminators), XML and YML

1289

1

 *** Regression tests ... maybe general cross validation

1290

1

 ** Misc: Focus detector, adaptive skin tracking

1291

1

 *** As discussed above

1292

1

 # Optimizations

1293

1

 ** .

1294

1

 # Documentation

1295

1

 ** .

1296

1

 # High GUI

1297

1

 ** .

1298

1

1299

1

_*OTHER*_

1300

1

1301

1

Summer of code

1302

1

1303

1

 * We already have not just a cascade, but a cascade-tree

1304

1

 * Gradient boosting trees(?) -- project?

1305

1

 * Optimize general boosting, not haar

1306

1

 * Have Mechanical Turk

1307

1

 * Image stitching

1308

1

 * HighGUI

1309

1

1310

1

New interface

1311

1

1312

1

 * Filtering functions in cv all pass test

1313

1

 ** SSE optimizations added for most important filtering functions

1314

1

 * Image warping next

1315

1

 * Usergroup glitch (old macro) posted to SVM

1316

1

1317

1

FLANN

1318

1

1319

1

 * Investigation is continuing by Anatoly

1320

1

 ** With new FLANN #2, the performance gap is smaller

1321

1

 *** Accuracy is the same

1322

1

 *** kd-tree in OpenCV is very fast, but not so accurate

1323

1

 *** FLANN, accuracy is high, but speed is slower.

1324

1

 **** We made it faster, with same accuracy, but don't know how to make it less accurate for more speed.

1325

1

1326

1

Extremely Random Trees

1327

1

1328

1

 * Is in. Needs sample code

1329

1

1330

1

MSER

1331

1

1332

1

 * Got it, stripped out other parts and compiled it.

1333

1

 * *Alexander* working on it.

1334

1

1335

1

3D Box

1336

1

1337

1

 * There was a bug, found it, corrected it.

1338

1

 * Sent email

1339

1

1340

1

1341

1

_*REPORTS*_

1342

1

1343

1

*Vadim*

1344

1

1345

1

# Debugged the new filtering functions (called by cvSmooth, cvFilter2D, cvErode, cvDilate, cvSobel). Now all the tests pass. Also, added SSE2 optimization to some important functions, like morphology and separable filters (like Sobel and Gaussian), so they should be much faster even if IPP is not available.

1346

1

# Started revision of image warping functions.

1347

1

# *Anatoly* continued investigation of various n-dim search algorithms. In the latest version of benchmark and with FLANN #2 the FLANN part runs faster, but still there is gap in speed-accuracy tradeoff. Also, the LSH part still does not give adequate results to be included into the benchmark. These 2 items are being investigated. *Anatoly* also ported the benchmark to Linux and sent it to *Marius*.

1348

1

1349

1

*Victor*

1350

1

1351

1

* See outlet detection for main report

1352

1

* 3D tracking: several bugs have been corrected, the new video is at ftp://argus-cv.dnsalias.org/3DTrackDemo/logitec_demo.avi, login willow, password awgarage.

1353

1

1354

1

1355

1

*_Action Items_*

1356

1

1357

1

*_NEW_*

1358

1

1359

1

*Gary*

1360

1

1361

1

 * Put up GSofC projects

1362

1

 * Job application page

1363

1

 * Put up application

1364

1

 * Ask *Marius* to interact with *Anatoly* on FLANN

1365

1

1366

1

*Vadim*

1367

1

1368

1

 * Put up GSofC projects

1369

1

1370

1

*Victor*

1371

1

1372

1

 * Put up GSofC projects

1373

1

1374

1

*_FROM LAST TIME_*

1375

1

1376

1

*Gary*

1377

1

1378

1

 * -Make change log wiki page, and send to *Vadim*-

1379

1

 * -Make OpenCV Montly Newsletter and send to *Vadim*-

1380

1

 * Ask J-Y again about mask image uses/image stitching

1381

1

 * -Find JD's code and send to *Vadim*-

1382

1

 * -Send email to *Radu* to point to mesh code/techniques-

1383

1

 * Calendor to Andrew Ng

1384

1

1385

1

*Vadim*

1386

1

1387

1

 * Put up change log wiki page and monthly page after *Gary* set's up

1388

1

1389

1

*Victor*

1390

1

1391

1

 * Do new video of box

1392

1

1393

1

h2. 2009-03-17

1394

1

1395

1

*_Agenda_*

1396

1

1397

1

[No meeting next week, I won't be in]

1398

1

1399

1

Google Summer of Code

1400

1

1401

1

HOG and SPIN

1402

1

1403

1

3D rotation + Translation of stereo images

1404

1

1405

1

Updates

1406

1

1407

1

 * New interface

1408

1

1409

1

"Monthly announce":http://opencv.willowgarage.com/wiki/OpenCV%20Monthly, "Change Logs":http://opencv.willowgarage.com/wiki/OpenCV%20Change%20Logs

1410

1

1411

1

FLANN Benchmarks

1412

1

1413

1

3D Box

1414

1

1415

1

3D capture

1416

1

1417

1

*_Minutes_*

1418

1

1419

1

"OpenCV Summer Release":http://opencv.willowgarage.com/wiki/OpenCV%20Ideas%202009

1420

1

1421

1

 # General Structure:

1422

1

 ** Decisions on library structure (Google's request, separate packages as is ...)

1423

1

 *** Provide way to build OpenCV statically, then size varies with binary size

1424

1

 # New Features

1425

1

 ** C++ interface

1426

1

 *** cxcore

1427

1

 *** cv  -- start with filtering functions, resize, optical flow, object detection (2 weeks)

1428

1

 ** Better python

1429

1

 *** *Vadim*, *Alex* ...

1430

1

 *** C++ will take another month but can improve python in parallel

1431

1

 ** 2D features C-descriptor, HoG Boundary Fragments ...

1432

1

 *** C-descriptor ... resolved issues with Michael, *Vadim* understands code more deeply (next week)

1433

1

 **** No re-implementation yet. Code based on randomized tree, but specialized version.

1434

1

 *** Boundary fragments ... working on focus.

1435

1

 *** Get contributions of 2D features externally

1436

1

 *** MSER ... (*Gary* has code)

1437

1

 *** HoG

1438

1

 *** Star is already in and in for ROS build

1439

1

 ** Alpha channel, alternatives to contours RLE

1440

1

 *** New alpha transparency (per pixel) added

1441

1

 *** Introduce functions with advance ways of blending

1442

1

 *** Mask channels

1443

1

 *** Optical flow

1444

1

 ** 3D features (Plane fit, PFH, FPFH from Radu), 3D Delaunay, 3D convex hull, mesh generation

1445

1

 *** SPIN -almost- done

1446

1

 *** *Radu* is organizing his code into a library ... maybe we can just lift this

1447

1

 ** 3D capture  http://www.3dsom.com/index.html

1448

1

 ** VSLAM

1449

1

 *** Wait on what *Kurt* and *James* get together

1450

1

 ** Image stitching

1451

1

 *** will ask Jean-Yves for pointers

1452

1

 ** Optical flow

1453

1

 *** Move code derived form former work in OpenCV

1454

1

 ** MLL

1455

1

 *** Extreme trees are coming

1456

1

 *** Reduce memory footprint

1457

1

 *** Need to work on input and output of data in CSV (or other simple deliminators), XML and YML

1458

1

 *** Regression tests ... maybe general cross validation

1459

1

 ** Misc: Focus detector, adaptive skin tracking

1460

1

 *** As discussed above

1461

1

 # Optimizations

1462

1

 ** .

1463

1

 # Documentation

1464

1

 ** .

1465

1

 # High GUI

1466

1

 ** .

1467

1

1468

1

_*OTHER*_

1469

1

1470

1

GSoC

1471

1

1472

1

 * Already received an application, but Google decides who's in on the 18th

1473

1

 * I've been working on some stereo data collection for people detection

1474

1

 ** Also for object detection too

1475

1

1476

1

HOG

1477

1

1478

1

 * 2D feature

1479

1

1480

1

SPIN

1481

1

1482

1

 * Code and

1483

1

 * Demo reads vrml

1484

1

 * Constructs oct tree C++

1485

1

 * Has compute normal points separate functions (takes oct tree)

1486

1

 * Then computes spin image

1487

1

1488

1

3D ROTATION TO NEW IMAGE

1489

1

1490

1

 * 3d perspective transform

1491

1

 * perspective projection

1492

1

 * for each point, find generalized interpolation

1493

1

 ** Computer weighted sum of K nearest points

1494

1

 ** To search in projected view, use quad tree or regular grid

1495

1

1496

1

*Vadim* update

1497

1

1498

1

# Made another few corrections in filtering functions to make the API a bit more convenient. Updated filtering-depdendent functions (that used not cvSobel, cvFilter2D etc., but the internal API): SURF, Snakes, CornerMinEigenVal etc. Now the tests for these functions pass too. Here are some performance numbers (on Core2Duo #6GHz laptop, assuming that performance of C code has not been changed much comparing to the old version of OpenCV, this is how much faster the summer version will be comparing to #0 and #1pre).

1499

1

# Started the work on updating image warping functions. cvResize & cvWarpAffine have been rewritten and optimized using SSE2 intrinsics. The code is now being debugged.

1500

1

1501

1

*_FLANN:_*

1502

1

1503

1

# *Anatoly* finished benchmarking FLANN and OpenCV's KD-tree.

1504

1

 * Below is the table (M1 and M3 mean the same thing as in previous report:

1505

1

 ** M1 - ratio of the distance to the found neighbor to the average distance to points in DB. The smaller - the better.

1506

1

 ** M3 - normalized to standard deviation difference between average distance and best distance. >= 2-2.5 is already very good. The larger - the better.

1507

1

1508

1

1509

1

 * _Conclusion_

1510

1

 ** Current conclusion is that OpenCV's kd-tree is comparable to FLANN in both speed and accuracy (by varying parameters we can get either better accuracy or better performance).

1511

1

 ** *Anatoly* now started working on HoG and the framework for model matching using spin-images.

1512

1

1513

1

3D Box

1514

1

1515

1

 * Sent video of tracking another box

1516

1

 * Put in local SVN

1517

1

1518

1

Application area

1519

1

1520

1

 * Under trunk put OpenCV Apps

1521

1

1522

1

3D Reconstruction

1523

1

1524

1

 * Nothing

1525

1

1526

1

*_Action Items_*

1527

1

1528

1

*_NEW_*

1529

1

1530

1

*Gary*

1531

1

1532

1

 * Talk to *Radu* about 3D rotation of points and create an image

1533

1

 * *Gary* to send *Anatoly* stereo data for trying HOG and SPIN

1534

1

 * Send gaussian curvature paper

1535

1

 * *Kurt* to send paper on fast plane fit

1536

1

 * Histogram of Gradient link

1537

1

1538

1

*Vadim*

1539

1

1540

1

 * *Anatoly* -- prioritize HOG, SPIN

1541

1

 ** Fast plane fit

1542

1

1543

1

*Victor*

1544

1

1545

1

 * Set up the first OpenCV appliation: 3D box

1546

1

 ** Video of it working

1547

1

 ** Raw video that spits out marked object

1548

1

1549

1

*_FROM LAST TIME_*

1550

1

1551

1

*Gary*

1552

1

1553

1

 * Put up GSofC projects

1554

1

 * Job application page

1555

1

 * Put up application

1556

1

 * Ask *Marius* to interact with *Anatoly* on FLANN

1557

1

1558

1

*Vadim*

1559

1

1560

1

 * Put up GSofC projects

1561

1

1562

1

*Victor*

1563

1

1564

1

 * Put up GSofC projects

1565

1

1566

1

h2.  2009-03-31

1567

1

1568

1

*_Agenda_*

1569

1

1570

1

Interface progress

1571

1

1572

1

 * Python machine learning

1573

1

1574

1

Updates for

1575

1

1576

1

 * Accomplishments

1577

1

 * "Monthly announce":http://opencv.willowgarage.com/wiki/OpenCV%20Monthly, "Change Logs":http://opencv.willowgarage.com/wiki/OpenCV%20Change%20Logs

1578

1

 * Contributors credit page?

1579

1

1580

1

Spin and HOG progress

1581

1

1582

1

 * How was HOG done?

1583

1

1584

1

Trying 3D capture code

1585

1

1586

1

Point cloud library

1587

1

1588

1

Application areas

1589

1

1590

1

New HighGUI

1591

1

1592

1

Improving the machine learning interface(?)

1593

1

1594

1

 * There is also the issue of a new classifier design for robotics that needs the following capabilities:

1595

1

 *** Needs to do one shot learning (but can refine later)

1596

1

 *** Needs to be able to combine separately learned objects in a sub-linear way

1597

1

 *** Needs the ability to express some structure in the data, particularly spatial or geometric structure.

1598

1

1599

1

*_Minutes_*

1600

1

1601

1

"OpenCV Summer Release":http://opencv.willowgarage.com/wiki/OpenCV%20Ideas%202009

1602

1

1603

1

 # General Structure:

1604

1

 ** Decisions on library structure (Google's request, separate packages as is ...)

1605

1

 *** Provide way to build OpenCV statically, then size varies with binary size

1606

1

 # New Features

1607

1

 ** C++ interface

1608

1

 *** cxcore

1609

1

 *** cv  -- start with filtering functions, resize, optical flow, object detection (2 weeks)

1610

1

 ** Better python

1611

1

 *** *Vadim*, *Alex* ...

1612

1

 ** 2D features C-descriptor, HoG Boundary Fragments ...

1613

1

 *** C-descriptor ... resolved issues with Michael, *Vadim* understands code more deeply (next week)

1614

1

 **** No re-implementation yet. Code based on randomized tree, but specialized version.

1615

1

 *** Boundary fragments ... working on focus.

1616

1

 *** Get contributions of 2D features externally

1617

1

 *** MSER ... (*Gary* has code)

1618

1

 *** HoG

1619

1

 *** Star is already in and in for ROS build

1620

1

 ** Alpha channel, alternatives to contours RLE

1621

1

 *** New alpha transparency (per pixel) added

1622

1

 *** Introduce functions with advance ways of blending

1623

1

 *** Mask channels

1624

1

 *** Optical flow

1625

1

 ** 3D features (Plane fit, PFH, FPFH from Radu), 3D Delaunay, 3D convex hull, mesh generation

1626

1

 *** SPIN -almost- done

1627

1

 *** *Radu* is organizing his code into a library ... maybe we can just lift this

1628

1

 ** 3D capture  http://www.3dsom.com/index.html

1629

1

 ** VSLAM

1630

1

 *** Wait on what *Kurt* and *James* get together

1631

1

 ** Image stitching

1632

1

 *** will ask Jean-Yves for pointers

1633

1

 ** Optical flow

1634

1

 *** Move code derived form former work in OpenCV

1635

1

 ** MLL

1636

1

 *** Extreme trees are coming

1637

1

 *** Reduce memory footprint

1638

1

 *** Need to work on input and output of data in CSV (or other simple deliminators), XML and YML

1639

1

 *** Regression tests ... maybe general cross validation

1640

1

 ** Misc: Focus detector, adaptive skin tracking

1641

1

 *** As discussed above

1642

1

 # Optimizations

1643

1

 ** .

1644

1

 # Documentation

1645

1

 ** .

1646

1

 # High GUI

1647

1

 ** .

1648

1

1649

1

_*OTHER*_

1650

1

1651

1

*Vadim*:

1652

1

1653

1

 # Revised, converted to the new API and tested:

1654

1

 **** all the image warping functions (Resize, WarpAffine, WarpPerspective, Remap & Undistort), with SSE2 optimization the performance improves up to 50-100%.

1655

1

 **** integral images

1656

1

 **** fixed-level threshold and adaptive threshold + SSE2 optimization

1657

1

 **** functions for background statistics accumulation

1658

1

 **** For the previously added median blur function added SSE2 optimization for 3x3 and 5x5 kernels, up to x10 speedup.

1659

1

 # Fixed haartraining app. problem (reported by Caroline Pantofaru) by reverting couple of files to the versions from OpenCV #0.

1660

1

 # Modified spin images to use the new C++ matrix class. The updated code is in SVN.

1661

1

 # Integrated MSER code by Liu Liu. For now the C interface is retained.

1662

1

 # *Anatoly* implemented function for extracting HoG features based on the contributed by Alberto Orlandi code. At new version is also ~50% faster. Some further optimization is possible. The code is in SVN now.

1663

1

 # Started working with the dataset of IKEA objects. HoG features did not provide any reasonable output (e.g. comparing to SURF), because the object outlines are very simple and HoG seems to require a classifier on top of it. Now trying to run spin images.

1664

1

1665

1

 * Next for API is optical flow

1666

1

 ** Pyramid and adding

1667

1

 ** Dense optical flow)

1668

1

 ** High Gui

1669

1

 * MLL

1670

1

 ** Extremely random trees Importance Variables

1671

1

 ** Input interface: to csv and loaded with standard function in SVN tomorrow

1672

1

1673

1

 * ROS stable this week.

1674

1

 ** latest_tested_snapshot is the SVN flag for ROS

1675

1

1676

1

Stereo data

1677

1

1678

1

 * On mechanical turk

1679

1

 * Radu, has developed a plane finder interface

1680

1

1681

1

Nearest neighbor

1682

1683

1684

1685

Spin features

1686

1

1687

1

 * Tried on synthetic data and vmrl files

1688

1

 * Missing elements vs Andrew Johnson features

1689

1

 ** Using geometric constraints if occluded object

1690

1

 * API 3D point cloud from oct tree

1691

1

 ** Pass set of points into spin function

1692

1

 ** List of 3D locations to compute spin images at

1693

1

 ** Computes normals

1694

1

 ** Pass pts and normals to a function that computes spin images

1695

1

 *** Support is the radius of included points

1696

1

 * Memory and time hogs -- hard to optimize

1697

1

1698

1

HOG

1699

1

1700

1

 * Anatoli tried HOG in a "Haar style" manner

1701

1

 * HOG features compared in matching set of points

1702

1

1703

1

Lighting

1704

1

1705

1

 * Want to start getting in specularity and lighting direction detectors

1706

1

 * Train by by rendering

1707

1

1708

1

Stanford feature

1709

1

1710

1

 * Sub patches from mean

1711

1

1712

1

3D capture code

1713

1

1714

1

 * Start *Anatoly* on 3D capture

1715

1

1716

1

*_Action Items_*

1717

1

1718

1

*_NEW_*

1719

1

1720

1

*Gary*

1721

1

1722

1

*Vadim*

1723

1

1724

1

 * Check kdd trees on small vectors

1725

1

1726

1

*Victor*

1727

1

1728

1

*_FROM LAST TIME_*

1729

1

1730

1

*Gary*

1731

1

1732

1

 * Talk to *Radu* about 3D rotation of points and create an image

1733

1

 * *Gary* to send *Anatoly* stereo data for trying HOG and SPIN

1734

1

 ** Look in

1735

1

 ** http://vault.willowgarage.com/wgdata1/vol1/IKea_Related/Class

1736

1

 ** For file Objects_kjlKJL.tar.gz (0.5G)

1737

1

 * Send gaussian curvature paper

1738

1

 * *Kurt* to send paper on fast plane fit

1739

1

 * Histogram of Gradient link

1740

1

1741

1

*Vadim*

1742

1

1743

1

 * *Anatoly* -- prioritize HOG, SPIN

1744

1

 ** Fast plane fit

1745

1

1746

1

*Victor*

1747

1

1748

1

 * Set up the first OpenCV appliation: 3D box

1749

1

 ** Video of it working

1750

1

 ** Raw video that spits out marked object

1751

1

1752

1

h2.  2009-04-07

1753

1

1754

1

*_Agenda_*

1755

1

1756

1

 * We need to update:

1757

1

 ** Accomplishments

1758

1

 ** "Monthly announce":http://opencv.willowgarage.com/wiki/OpenCV%20Monthly

1759

1

 ** "Change Logs":http://opencv.willowgarage.com/wiki/OpenCV%20Change%20Logs Can we get new functions listed here?

1760

1

 ** Contributors credit page?

1761

1

 ** Update what people are working on.

1762

1

 * *Vadim*

1763

1

 ** Stereo block match at two scales.  Progress?

1764

1

 ** kd_tree tests for small vectors?

1765

1

 ** New interface progress

1766

1

 ** 3D Capture -- software in? Testing?

1767

1

 ** Python interface based on numpy matrices?

1768

1

 ** Optical flow code?

1769

1

 * *Victor*

1770

1

 ** Code for 3D box tracking

1771

1

 ** Chamfer matching

1772

1

 *** 2D

1773

1

 *** 3D?

1774

1

 ** Machine learning reading csv

1775

1

 *** Other MLL improvements

1776

1

1777

1

*_Minutes_*

1778

1

1779

1

*Victor*

1780

1

1781

1

 * 3D box code (put into OpenCV application section)

1782

1

 * Chamfer matching

1783

1

 ** Used in outlet matching -- works well for orange matching

1784

1

 *** Many false positives in high texture

1785

1

 *** Works worse on white on white because edges are unreliable

1786

1

 *** Useful for European outlets

1787

1

 *** Edge template better, but depends on edge fidelity

1788

1

 * MLL

1789

1

 ** a major commit to opencv svn.

1790

1

 *** The new memory-efficient structure for tree-based classifiers has been implemented. The new memory limitations are still under investigation (because of the recent changes in cxcore).

1791

1

 *** About 2X saving for old version of OpenCV, looking at new version

1792

1

 *** 10,000 samples, 1000 variables.

1793

1

 *** Doing new tests (*Maria*)

1794

1

1795

1

Application section

1796

1

1797

1

 * Same svn link, but put opencv_extra

1798

1

 * For example https://opencvlibrary.svn.sourceforge.net/svnroot/opencvlibrary/trunk/opencv_extra/3d/createsamples3d/

1799

1

1800

1

*Vadim*

1801

1

1802

1

 * Stereo starting tommorrow

1803

1

 * kd-tree on small vectors -- will send table

1804

1

1805

1

Activity results:

1806

1

1807

1

 # Converted to the new API and optimized (by factor of 2) pyramid functions.

1808

1

 # Integrated Fuzzy MeanShift Tracker by Farhad Dadgostar (the same person who contributed adaptive skin detector).

1809

1

 # Integrated Farnerback dense optical flow.

1810

1

 # *Anatoly* started investigation of 3DSom software for 3D reconstruction. A brief report has been sent.

1811

1

1812

1

Plans for this week:

1813

1

1814

1

 # Purchase some lighting setup for better background isolation. We have found some good possible setups in the internet already.

1815

1

 # Finish optical flow intergration (other dense optical flow algorithm and better sparse optical flow).

1816

1

 # Implement adaptive-window variant of stereo correspondence algorithm

1817

1

 * Issues: we still have not received the license update from 3DSom.

1818

1

1819

1

Python

1820

1

1821

1

 * Can create numpy array and call OpenCV function

1822

1

 * C-type tool in more use

1823

1

 * Supports overload

1824

1

1825

1

Optical Flow

1826

1

1827

1

 * Forward back algorithm

1828

1

 * Have another dense algorithm almost ready, will integrate

1829

1

 * Improve sparse optical flow with color and image gradients

1830

1

 * About 1 week more work (delayed for stereo)

1831

1

1832

1

Contours and Object detection

1833

1

1834

1

 * Need C++ interface

1835

1

 * Contours

1836

1

 ** 2 week working time (data structure issues)

1837

1

 * Object detection

1838

1

 ** Probably 1 week

1839

1

1840

1

HighGUI and MLL

1841

1

1842

1

 * Can be converted in interface

1843

1

 * HighGUI fast

1844

1

 * MLL ~ 1 month

1845

1

1846

1

External code, improvement in

1847

1

1848

1

The details of the implementation can be found here:

1849

1

1850

1

http://www.springerlink.com/content/p32p311820r63k68/

1851

1

1852

1

and one of the related videos also is here:

1853

1

1854

1

http://www.youtube.com/watch?v=gVPCrQBKKqE&feature=channel_page

1855

1

1856

1

"OpenCV Summer Release":http://opencv.willowgarage.com/wiki/OpenCV%20Ideas%202009

1857

1

1858

1

 # General Structure:

1859

1

 ** Decisions on library structure (Google's request, separate packages as is ...)

1860

1

 *** Provide way to build OpenCV statically, then size varies with binary size

1861

1

 # New Features

1862

1

 ** C++ interface

1863

1

 *** cxcore

1864

1

 *** cv  -- start with filtering functions, resize, optical flow, object detection (2 weeks)

1865

1

 ** Better python

1866

1

 *** *Vadim*, *Alex* ...

1867

1

 ** 2D features C-descriptor, HoG Boundary Fragments ...

1868

1

 *** C-descriptor ... resolved issues with Michael, *Vadim* understands code more deeply (next week)

1869

1

 **** No re-implementation yet. Code based on randomized tree, but specialized version.

1870

1

 *** Boundary fragments ... working on focus.

1871

1

 *** Get con

Login	Password