moothes / itsd-pytorch Goto Github PK

Hi, @moothes.

During my check of your code, I have a question about your ResNet50 pkl file.

I understand the pkl file is for your whole model, so it include the pretrain ResNet.
But when I try to find the parameter in your pkl file, the ResNet part in your pkl file has different value compared with official pre-train model from pytorch. They have same value only for BatchNorm layers(i.e. running_mean, running_var, weight, bias).

as you can see, conv weights are not the same.
I confirm both pkl file and pth file are downloaded on 15 Feb, 2021, using your link in github, so they are the latest.

to me, if you freeze the ResNet50, then both conv_weight and BN should be the same as pytorch pre-train model.
if we train the pre-train ResNet50, then both conv_weight and BN should not be the same as pytorch pre-train model,
but here, conv_weight not the same, BN parameters are the same...
Do you know what is the reason?

你好，
感谢你所做的贡献，我想请问一下训练和测试的时候，数据集路径应该怎么设置，因为我设置以后，总给我报错‘系统找不到指定的路径。: '../dataset/DUTS-TR/images'’，按照这个报错路径设置了也还是报这个错误。
谢谢

I am very interested in your work. I runned your codes, but i met some problems. Could you help me ?
/src/Metrics.py
TPs = [torch.sum(torch.sum((_Y >= threshold) & (self.Y), -1), -1).float() for threshold in self.thresholds]
RuntimeError: Expected object of scalar type Bool but got scalar type Byte for argument #2 'other'

hi,
First, thank you for sharing your excellent work.
However, I have some issues.
I would like to ask what GPU you used when testing, including the image size used when testing the fps of other models. Is it convenient to tell? Thank you

Hi,
Is it possible to upload the predicted saliency maps into Google Drive?
I can't download them.

Hello @moothes thank you for your great work. I have a problem with the link to download the result and your pretrained model. I can't download your data that uploaded in baidu. I think your link is not working. Could you reupload it?
-Thank you-

Missing key(s) in state_dict

@moothes May I know what is the purpose of having these weights = [0.1, 0.3, 0.5, 0.7, 0.9, 1.5] in Loader.py? How those weights are determined?

First of all, thanks for your amazing work as well as sharing the necessary resources to encourage the reproducibility of your work. However, I have been encountering some issues to even executing the testing with your pretrained model by following the exact instruction as provided in the readme. I was first attempted with a newer version of PyTorch and TorchVision but later on decided to setup a new virtual environment in order to install the right versions as stated. However, such attempt did not guarantee a smooth execution and I was having the following issue: AttributeError: module 'torch.nn' has no attribute 'SyncBatchNorm'. After some searches, the recommended workaround is to upgrade the PyTorch version from 1.0 to 1.1 and it seemed to be working to solve the SyncBatchNorm issue. Just wondering have you encountered the same issue as I did since the exact same versions of the PyTorch and TorchVision were used?

Another issue that I encountered is as follows: RuntimeError: Error(s) in loading state_dict for baseline: Missing key(s) in state_dict: "total_ops", "total_params", " ... and it seems the problem is coming from self.Model.load_state_dict(torch.load(L.mpath, map_location='cpu')) in the Experiment.py. Appreciate if you could provide me with more details on how such issue could be tackled. Thank you.

I run the train process, but it outputs errors as follows:

loading the settings
Backbone: resnet, Using Gpu: 1

Traceback (most recent call last):
File "train.py", line 10, in
train()
File "train.py", line 4, in train
L = Loader('train')
File "/ITSD-pytorch/src/Loader.py", line 54, in init
self.loading(opt)
File "/ITSD-pytorch/src/Loader.py", line 129, in loading
self.Model = baseline(self.model, self.channel)
TypeError: 'module' object is not callable

Please help me. How to solve it? Thx.