klrc / racnn-pytorch Goto Github PK

Hi @klrc ,

First of all great repo!!
I have been trying to replicate the results of this paper. Till now, I am stuck at 68% top-1 accuracy. I saw you got 69 top-1 accuracy at 23 epochs. Did you get chance to complete the training and get better results?

Please Let me know thanks!

Thank you very much for your code; it has been a great help to me! However, I have a small question I'd like to ask, if I can still receive your response.

Firstly: After I call the save_img function, the saved image appears as a blank gray image (Image 1). I debugged the issue and it seems to be related to the tensor_to_img function during the value conversion process (since I can display the magnified image using plt.imshow() with colors, as shown in Image 2). However, I'm unsure which step is causing the current problem. Do you have any suggestions?

Secondly: I have a minor question about an image similar to my Image 2. Why are there gradient-like color blocks on top? Even when I initially load the original image using plt.imshow(), it also contains these color blocks, rather than the most original image.
Image 1：

Image 2：

Note: I've replaced the dataset with one related to medical images.

Hi, @klrc
I just started learning RA-CNN, and I have a question for you. Here is your result on the testset:

[2019-12-31 20:06:50] :: Testing on test set ...
[2019-12-31 20:07:10] Accuracy clsf-0@top-1 (201/725) = 79.95050%
[2019-12-31 20:07:10] Accuracy clsf-0@top-5 (201/725) = 94.61634%
[2019-12-31 20:07:10] Accuracy clsf-1@top-1 (201/725) = 74.25743%
[2019-12-31 20:07:10] Accuracy clsf-1@top-5 (201/725) = 91.39851%
[2019-12-31 20:07:10] Accuracy clsf-2@top-1 (201/725) = 74.62871%
[2019-12-31 20:07:10] Accuracy clsf-2@top-5 (201/725) = 90.71782%

I wonder why the accuracy of clsf-0, clsf-1, clsf-2 decreases in order? Shouldn't clsf-2 be a more fine-grained region compared with clsf-1 and clsf-0?
I got the same results on my datasets. If you can help me with this question, I would appreciate it very much! Thank you!

Thanks for making this project, It's convenient for me to implement RACNN in PyTorch.😄
However, When I load a trained model, I found the error below:

(pytorch) root@server:~/RACNN-PYTORCH_orchids# python test.py
Traceback (most recent call last):
  File "test.py", line 12, in <module>
    model = torch.load("build/racnn_mobilenetv2_cub200-e40se401635929775.pt")
  File "/home/user/anaconda3/envs/pytorch/lib/python3.8/site-packages/torch/serialization.py", line 607, in load
    return _load(opened_zipfile, map_location, pickle_module, **pickle_load_args)
  File "/home/user/anaconda3/envs/pytorch/lib/python3.8/site-packages/torch/serialization.py", line 882, in _load
    result = unpickler.load()
  File "/home/user/anaconda3/envs/pytorch/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1130, in __getattr__
    raise AttributeError("'{}' object has no attribute '{}'".format(
AttributeError: 'RACNN' object has no attribute '__echo_apn'

After I did some researches, I guess the problem is the typo of these two lines:

If I replace model.state_dict with model.state_dict(), the probelm was solved.
Hope you can edit your code, thanks for your hard work!