search

sud0301 / semisup-semseg

136 stars 30 forks source link

IOU resluts problem about train_s4GAN.py #15

Open tsrobcvai opened 3 years ago

tsrobcvai commented 3 years ago

Hello, thank you for your outstanding research work! I encountered such a problem. When I ran train_s4GAN.py (labeled-ratio 0.125/PASCAL-VOC Dataset/number_steps 35000), the best IOU I got was about 0.65, which is quite different from the 0.698 in the paper . The following is my visualization, Can you give me some suggestions, thank you! %AQ5$STJ_E1YZFS~9X6DG J 7JZ7PHW(6 QQH0CG_)_EA15

sud0301 commented 3 years ago

Hi, Just a reminder, the paper reports performance of 69.8 mIOU with COCO+ImageNet pre-training. We report a performance of 65.4 in the paper when the network is pretrained only on ImageNet. Can you please verify whether you are using only ImageNet or ImageNet+COCO for pretraining the network.

tsrobcvai commented 3 years ago

Hi, Just a reminder, the paper reports performance of 69.8 mIOU with COCO+ImageNet pre-training. We report a performance of 65.4 in the paper when the network is pretrained only on ImageNet. Can you please verify whether you are using only ImageNet or ImageNet+COCO for pretraining the network.

Yes it is! I just used only ImageNet for pretraining the network what you provided here，I thought only one kind of pre-training was needed (I'm a rookie). Could you tell me where I can find it ImageNet+COCO pretrained model？And I have one more question，from the VAl_loss of my training, there is a problem of overfitting. Does this mean that the model can get better results after careful Tuning parameters? thank you！