CVI-SZU / CLIMS

[CVPR 2022] CLIMS: Cross Language Image Matching for Weakly Supervised Semantic Segmentation
MIT License
124 stars 12 forks source link

Hi,should I create a directory cam-baseline-voc12? #6

Closed yangxinhaosmu closed 2 years ago

yangxinhaosmu commented 2 years ago

Should I create a directory cam-baseline-voc12 and put res50_cam.pth in it? Is the res50_cam.pth pre-trained on ImageNet by your team? What is the difference between this res50_cam.pth and net/resnet50_cam.py trained by myself? If I run the run_sample.py command line, am I just train your .pth more time(epoch) or I train the model from scratch? Sorry to ask such many easy questions, I am the beginner of WSSS. Thank you so much!

Sierkinhane commented 2 years ago

Hi,

  1. I provided res50_cam.pth which was the pre-trained model on VOC2012. Actually, you can try to train such a model or directly use ours (create a directory called cam-baseline-voc12 then put the pretrained res50_cam.pth in it).
  2. resnet50_cam.py is the basic network (baseline cam). res50_cam.pth is the saved model of resnet50_cam.py
  3. run_sample.py will load the pretrained model to train clims. btw, welcome to star our repository
yangxinhaosmu commented 2 years ago

Hi,

  1. I provided res50_cam.pth which was the pre-trained model on VOC2012. Actually, you can try to train such a model or directly use ours (create a directory called cam-baseline-voc12 then put the pretrained res50_cam.pth in it).
  2. resnet50_cam.py is the basic network (baseline cam). res50_cam.pth is the saved model of resnet50_cam.py
  3. run_sample.py will load the pretrained model to train clims. btw, welcome to star our repository
yangxinhaosmu commented 2 years ago

Hi,

  1. I provided res50_cam.pth which was the pre-trained model on VOC2012. Actually, you can try to train such a model or directly use ours (create a directory called cam-baseline-voc12 then put the pretrained res50_cam.pth in it).
  2. resnet50_cam.py is the basic network (baseline cam). res50_cam.pth is the saved model of resnet50_cam.py
  3. run_sample.py will load the pretrained model to train clims. btw, welcome to star our repository

1.感谢,我目前的理解是先用res50_cam.pth的模型去初始化resnet50_clims里的CLIMS对吗? 2.还有一个问题就是如果我想CLIP的image encoder之前加一些模块(比如对特征表示做增强这种的)的话,可以在resnet50_clims里的CLIMS的stage后面直接加吗?这样的话还能用res50_cam.pth去初始化吗?

Sierkinhane commented 2 years ago

可以的,如果你添加了一些模块的话,也可以导入预训练的模型,只要把这里的strict=True换成strict=False https://github.com/CVI-SZU/CLIMS/blob/f37162c1fba3a0011600a6eadfe97fc16edd7edd/step/train_clims.py#L65

yangxinhaosmu commented 2 years ago

可以的,如果你添加了一些模块的话,也可以导入预训练的模型,只要把这里的strict=True换成strict=False

https://github.com/CVI-SZU/CLIMS/blob/f37162c1fba3a0011600a6eadfe97fc16edd7edd/step/train_clims.py#L65

我大概了解了,还有一个小小的问题,就是如果不导入res50_cam.pth的话,训练resnet50_clims大概需要多花多少时间呢?您有记录过吗?

Sierkinhane commented 2 years ago

最好导入一下预训练模型,效果会好些,总体训练时间是一样的应该是10epoch

yangxinhaosmu commented 2 years ago

最好导入一下预训练模型,效果会好些,总体训练时间是一样的应该是10epoch

OK! 感谢同学耐心的指导!!!祝申到理想的PHD!

Sierkinhane commented 2 years ago

感谢!