aiacademy: 深度學習 CNN Segmentation

Segmentation

要實現 Segmentation 靠: Fylly Convolutional Networks for Semantic Segmentation (FCN)

• Only Convolution layer and transposed convolution layer in the network architecture.

Approach

• To make the predicted resulte better, adding the skips to combine the final prediction layer with lower layer with finer stride.

Object Instance Segmentation

Object Instance Segmentation: Mask RCNN

Mask R-CNN

Mask RCNN

• Instance segmentation
• output: Masks
• Two Stage 的演算法
• Mask-RCNN 是 Faster-RCNN 的改良版
• Msdk_RCNN GitHub

Matching and co-segmentation (Optinal)