Embedded Files
1) Question:
8.1 What are the disadvantages of traditional CNN-based segmentation methods? 269
8.1 FCN 269
8.1.1 How has the FCN changed? 269
8.1.2 FCN network structure 270
8.1.3 Example of a full convolution network 271
8.1.4 Why is it difficult for CNN to classify pixels? 271
8.1.5 How do the fully connected and convolved layers transform each other? 272
8.1.6 Why can the input picture of the FCN be any size? 272
8.1.7 What are the benefits of reshaping the weight of the fully connected layer into a convolutional layer filter? 273
8.1.8 Deconvolutional Understanding 275
8.1.9 Skip structure 276
8.1.10 Model Training 277
8.1.11 FCN Disadvantages 280
8.2 U-Net 280
8.3 SegNet 282
8.4 Dilated Convolutions 283
8.4 RefineNet 285
8.5 PSPNet 286
8.6 DeepLab Series 288
8.6.1 DeepLabv1 288
8.6.2 DeepLabv2 289
8.6.3 DeepLabv3 289
8.6.4 DeepLabv3+ 290
8.7 Mask-R-CNN 293
8.7.1 Schematic diagram of the network structure of Mask-RCNN 293
8.7.2 RCNN pedestrian detection framework 293
8.7.3 Mask-RCNN Technical Highlights 294
8.8 Application of CNN in Image Segmentation Based on Weak Supervised Learning 295
8.8.1 Scribble tag 295
8.8.2 Image Level Marking 297
8.8.3 DeepLab+bounding box+image-level labels 298
8.8.4 Unified framework 299
2) Answer:
P
References:
Page updated
Google Sites
Report abuse