Taylor-X76

Results 5 comments of Taylor-X76

guide to use pre-trained model

loda .pth in model.init_weights() like this model.init_weights(pretrained = "/home1/xmy/DCL/HRNet_Image_Classification/experiments/hrnetv2_w18_imagenet_pretrained.pth")

A solution: HRNet Backbone Adopt_different_blocks_bug(BASIC//BOTTLENECK)

![HRNet_v2不同block_yaml文件](https://user-images.githubusercontent.com/51960362/91817261-d3f42180-ec67-11ea-9cbb-68e4dcb2bd7e.png) ![HRNet_v2不同block_bug](https://user-images.githubusercontent.com/51960362/91817363-da829900-ec67-11ea-9540-1b972e29ad49.png) ![HRNet_v2不同block_改正](https://user-images.githubusercontent.com/51960362/91817562-e66e5b00-ec67-11ea-952e-5dfd8ca15b58.png) ![HRNet_v2不同block_bug代码更改](https://user-images.githubusercontent.com/51960362/91817597-e8d0b500-ec67-11ea-838d-62140e1af0a2.png)

A solution: HRNet Backbone Adopt_different_blocks_bug(BASIC//BOTTLENECK)

`# ############################# Modified by Mingyang ############################################################## # ##################### Adopt_different_blocks_bug(BASIC//BOTTLENECK) ############################################# # x_list = [] # for i in range(self.stage3_cfg['NUM_BRANCHES']): # if self.transition2[i] is not None: # x_list.append(self.transition2[i](y_list[-1])) # else: #...

Train from scratch

train 10 epoch，C3D的ACC也是1% 另外我改了loss = criterion(probs, labels) 不然ACC会nan

Is random flip accomplished only in testing set?

II think so too After two cv2.flip(...), the picture has no change.