Can not download the pre-trained model about fcos

[tianzhi0549]

@ZhengMengbin Please download the model here https://cloudstor.aarnet.edu.au/plus/s/k3ys35075jmU1RP/download.

[ZhengMengbin]

I use the pre-trained model from https://cloudstor.aarnet.edu.au/plus/s/k3ys35075jmU1RP/download, but it happens an another error:
'File "tools/train_net.py", line 175, in
main()
File "tools/train_net.py", line 168, in main
model = train(cfg, args.local_rank, args.distributed)
File "tools/train_net.py", line 54, in train
extra_checkpoint_data = checkpointer.load(cfg.MODEL.WEIGHT)
File "/home/zhengchenbin/FcosNet/FCOS/maskrcnn_benchmark/utils/checkpoint.py", line 65, in load
checkpoint = self._load_file(f)
File "/home/zhengchenbin/FcosNet/FCOS/maskrcnn_benchmark/utils/checkpoint.py", line 138, in _load_file
return load_c2_format(self.cfg, f)
File "/home/zhengchenbin/FcosNet/FCOS/maskrcnn_benchmark/utils/c2_model_loading.py", line 175, in load_c2_format
return C2_FORMAT_LOADER[cfg.MODEL.BACKBONE.CONV_BODY](cfg, f)
File "/home/zhengchenbin/FcosNet/FCOS/maskrcnn_benchmark/utils/c2_model_loading.py", line 170, in load_resnet_c2_format
state_dict = _rename_weights_for_resnet(state_dict, stages)
File "/home/zhengchenbin/FcosNet/FCOS/maskrcnn_benchmark/utils/c2_model_loading.py", line 124, in _rename_weights_for_resnet
w = torch.from_numpy(v)
TypeError: expected np.ndarray (got str)'
It seems this pre-trained model do not the same model from https://dl.fbaipublicfiles.com/detectron/ImageNetPretrained/20171220/X-101-64x4d.pkl.

[tianzhi0549]

@ZhengMengbin Please add the following code

if 'weight_order' in k:
    continue

to

FCOS/maskrcnn_benchmark/utils/c2_model_loading.py

Line 122 in c51436a

# if 'fc1000' in k:

.
It will skip the key weight_order.

[ZhengMengbin]

It works, thank you very much for your reply! I will close this issue.

[ZhengMengbin]

I have another question, I download the trained model FCOS_X_101_64x4d_FPN_2x, its size is only 361.5M, but I use the pre-trained model from https://cloudstor.aarnet.edu.au/plus/s/k3ys35075jmU1RP/download(the pre-trained model's size is 667.7M), the saved model's size is 687M, why is the size gap so large?

[tianzhi0549]

@ZhengMengbin We remove all solver states for our released models.

[ZhengMengbin]

Is it convenient to provide a script file for removing all the solver states?

[tianzhi0549]

@ZhengMengbin We will add the script into our code when we have time. For now, you can use torch.load to load the model and then delete what you don't want. Next, you can use torch.save to save the pruned model.

[ZhengMengbin]

OK, special thanks, I will try it.

[ZhengMengbin]

I use the code as follows to remove all the solver states：

Using the above code, I got the same model size as the author.

[tianzhi0549]

@ZhengMengbin Thank you very much for posting it.