0.25
0.5
0.75
1.25
1.5
1.75
2
Convolutional Neural Networks and Computer Vision
Published on Aug 23, 201617009 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.
Related categories
Chapter list
Introduction to Convolutional Networks00:00
Overview00:18
Neural Net00:53
Convolutional Neural Networks01:07
Multistage Hubel-Wiesel Architecture01:35
Overview of Convnets02:39
Convnet Successes03:56
Application to ImageNet04:56
Goal05:49
Krizhevsky et al. [NIPS2012] 06:36
ImageNet Classification (2010 – 2015)07:33
Examples - 108:42
Examples - 209:00
Examples - 309:08
Using Features on Other Datasets09:44
Caltech 256 - 110:00
Caltech 256 - 210:24
The Details10:58
Components of Each Layer11:14
Filtering - 111:51
Filtering - 213:20
Filtering - 314:57
Non-Linearity - 217:30
Non-Linearity - 117:50
Pooling - 118:29
Pooling - 220:35
Role of Pooling20:57
Components of Each Layer23:25
Architecture 23:51
How to Choose Architecture24:16
How important is Depth25:59
Architecture of Krizhevsky et al. - 126:09
Architecture of Krizhevsky et al. - 227:25
Architecture of Krizhevsky et al. - 328:09
Architecture of Krizhevsky et al. - 428:13
Architecture of Krizhevsky et al. - 528:38
Tapping off Features at each Layer29:31
Translation (Vertical)29:52
Scale Invariance31:08
Rotation Invariance31:31
Very Deep Models (2)32:55
GoogLeNet vs Previous Models33:14
Google Inception model33:36
Very Deep Models (1)34:12
Residual Networks34:29
Visualizing Convnets - 137:49
Visualizing Convnets - 238:50
Projection from Higher Layers38:58
Details of Operation39:52
Unpooling Operation40:38
Layer 1 Filters41:22
Visualizations of Higher Layers41:42
Layer 1: Top-9 Patches42:22
Layer 2: Top-142:56
Layer 2: Top-9 Patches46:09
Layer 2: Top-946:30
Visualizing Convnets47:05
Google DeepDream47:47
Training Big ConvNets48:01
Evolution of Features During Training - 149:50
Evolution of Features During Training - 250:14
Normalization across Data50:41
Annealing of Learning Rate52:09
Automatic Tuning of Learning Rate?52:19
Local Minima?53:40
What about 2nd order methods?56:30
Saddle Point Perspective58:25
Improving Generalization59:36
Big Model + Regularize vs Small Model01:00:47
Fooling Convnets01:01:47
DropOut01:02:38
Other things good to know - 201:03:37
Other things good to know - 301:04:10
Other things good to know - 101:04:38
Other things good to know - 401:04:55
What if it does not work?01:05:06
Industry Deployment01:05:47
Labeled Faces in Wild Dataset01:06:23
Detection with ConvNets01:07:25
Two General Approaches01:07:39
Sliding Window with ConvNet01:08:28
Multi-Scale Sliding Window ConvNet - 101:08:45
Multi-Scale Sliding Window ConvNet - 201:09:04
OverFeat – Output before NMS01:09:20
Overfeat Detection Results01:10:06
R-CNN Approach01:10:24
Video Classification01:11:04
Action Recognition Results01:12:28
2D vs 3D Convnets01:12:58
Sport Classification Results01:13:19
Dense Scene Labeling - 101:14:09
Dense Scene Labeling - 201:14:52
Dense Scene Labeling - 301:15:03
Dense Scene Labeling - 401:15:06
Architecture01:15:29
Multi-Scale Convnets01:15:42
Eigen et al. architecture01:15:52
Use Appropriate Loss Functions01:16:07
Depths Comparison01:16:31
Surface Normals01:16:45
Scene Parsing01:16:51
Segmentation01:17:00
Denoising with ConvNets01:18:46
Deblurring with Convnets01:19:10
Inpainting with Convnets01:19:28
Removing Local Corruption01:19:38
Convnet + Structured Learning - 101:20:32
Convnet + Structured Learning - 201:21:12
Body Tracking01:21:16
Body Tracking: Part Detector01:21:48
Body Tracking: Spatial Model - 101:21:54
Body Tracking: Spatial Model - 201:22:23