0.25
0.5
0.75
1.25
1.5
1.75
2
Convolutional Neural Networks and Computer Vision
Published on 2016-08-2317008 Views
This talk will review Convolutional Neural Network models and the tremendous impact they have made on Computer Vision problems in the last few years.
Presentation
Introduction to Convolutional Networks00:00
Overview05:13:50
Neural Net14:53:53
Convolutional Neural Networks18:52:57
Multistage Hubel-Wiesel Architecture26:26:29
Overview of Convnets44:14:12
Convnet Successes65:48:48
Application to ImageNet82:14:13
Goal96:57:43
Krizhevsky et al. [NIPS2012] 110:14:21
ImageNet Classification (2010 – 2015)125:53:02
Examples - 1145:15:33
Examples - 2150:16:38
Examples - 3152:13:52
Using Features on Other Datasets162:14:05
Caltech 256 - 1166:53:58
Caltech 256 - 2173:35:02
The Details183:00:02
Components of Each Layer187:24:33
Filtering - 1197:36:58
Filtering - 2222:22:26
Filtering - 3249:17:42
Non-Linearity - 2291:48:47
Non-Linearity - 1297:22:54
Pooling - 1308:18:09
Pooling - 2343:13:06
Role of Pooling349:21:22
Components of Each Layer390:21:41
Architecture 397:35:07
How to Choose Architecture404:33:59
How important is Depth433:05:08
Architecture of Krizhevsky et al. - 1435:56:19
Architecture of Krizhevsky et al. - 2457:04:55
Architecture of Krizhevsky et al. - 3469:24:07
Architecture of Krizhevsky et al. - 4470:25:27
Architecture of Krizhevsky et al. - 5477:16:42
Tapping off Features at each Layer492:09:56
Translation (Vertical)497:50:20
Scale Invariance519:09:21
Rotation Invariance525:19:28
Very Deep Models (2)548:43:51
GoogLeNet vs Previous Models554:07:12
Google Inception model560:13:24
Very Deep Models (1)570:09:30
Residual Networks574:56:31
Visualizing Convnets - 1630:27:00
Visualizing Convnets - 2647:29:51
Projection from Higher Layers649:39:06
Details of Operation664:37:47
Unpooling Operation677:16:31
Layer 1 Filters689:28:18
Visualizations of Higher Layers695:15:01
Layer 1: Top-9 Patches706:10:09
Layer 2: Top-1715:41:47
Layer 2: Top-9 Patches769:22:09
Layer 2: Top-9775:14:49
Visualizing Convnets784:47:57
Google DeepDream796:35:07
Training Big ConvNets800:30:27
Evolution of Features During Training - 1830:49:51
Evolution of Features During Training - 2837:22:43
Normalization across Data844:52:33
Annealing of Learning Rate869:15:10
Automatic Tuning of Learning Rate?872:00:28
Local Minima?894:39:18
What about 2nd order methods?941:50:58
Saddle Point Perspective973:39:43
Improving Generalization993:29:59
Big Model + Regularize vs Small Model1013:12:21
Fooling Convnets1029:50:43
DropOut1043:59:27
Other things good to know - 21060:25:46
Other things good to know - 31069:26:45
Other things good to know - 11077:18:49
Other things good to know - 41082:03:35
What if it does not work?1085:10:01
Industry Deployment1096:26:37
Labeled Faces in Wild Dataset1106:25:34
Detection with ConvNets1123:38:00
Two General Approaches1127:45:56
Sliding Window with ConvNet1141:18:16
Multi-Scale Sliding Window ConvNet - 11145:55:04
Multi-Scale Sliding Window ConvNet - 21151:23:14
OverFeat – Output before NMS1155:48:25
Overfeat Detection Results1168:25:58
R-CNN Approach1173:25:48
Video Classification1184:40:53
Action Recognition Results1207:49:16
2D vs 3D Convnets1216:23:01
Sport Classification Results1222:00:38
Dense Scene Labeling - 11235:58:53
Dense Scene Labeling - 21247:56:32
Dense Scene Labeling - 31250:55:29
Dense Scene Labeling - 41251:46:42
Architecture1258:17:38
Multi-Scale Convnets1261:48:36
Eigen et al. architecture1264:32:32
Use Appropriate Loss Functions1268:52:07
Depths Comparison1275:18:30
Surface Normals1279:21:28
Scene Parsing1280:57:48
Segmentation1283:25:32
Denoising with ConvNets1312:49:13
Deblurring with Convnets1319:30:35
Inpainting with Convnets1324:43:19
Removing Local Corruption1327:16:03
Convnet + Structured Learning - 11342:17:38
Convnet + Structured Learning - 21353:28:21
Body Tracking1354:42:45
Body Tracking: Part Detector1363:21:30
Body Tracking: Spatial Model - 11365:06:42
Body Tracking: Spatial Model - 21373:18:13