MSRF-Net
MSRF-Net copied to clipboard
NoviceMAn-prog
I tried to train on the Kvasir dataset, but the performance was quite different from that of the paper. It would be better if the author could provide some parameters...
File "c:\Users\user\test\MSRF-Net\model.py", line 267, in msrf canny = cv2.Canny(np.asarray(inputs),10,100) what's the meaning of inputs? * can you upload the requirement file?
Can you make .py file of the predicted masks for each image in the test set? 
Can you make .py file of the predicted masks for each image in the test set? 
The last commit uncommented the following line, and the `inputs` variable is not defined. I didn't figure out the right argument for the `Canny` method. https://github.com/NoviceMAn-prog/MSRF-Net/blob/60e23734d94d340b9d89b64e7f9e2636b6f96c0f/model.py#L267
As in `msrf.py` and in your article, the connections for the following code should be: https://github.com/NoviceMAn-prog/MSRF-Net/blob/60e23734d94d340b9d89b64e7f9e2636b6f96c0f/model.py#L234-L236 ```python n13,n23 = RDDB(n13,n23,32,64,16) n33,n43 = RDDB(n33,n43,128,256,64) ```
Hello @NoviceMAn-prog, please delete `msrf.py` from your repo, it was the first network file that was divided later into separated files for each part of your code. Also, have more...
This PR fixes some images left in batch training when the following condition is never met due to `float` to `int` conversion. https://github.com/NoviceMAn-prog/MSRF-Net/blob/60e23734d94d340b9d89b64e7f9e2636b6f96c0f/train.py#L103 Also, used `tqdm` dependence that was never...
In Up3, the tranpose conv has kernel size (1,1). From what i know this is quite uncommon, is there a reason why this isn't (2,2) or (4,4) ? https://github.com/NoviceMAn-prog/MSRF-Net/blob/60e23734d94d340b9d89b64e7f9e2636b6f96c0f/model.py#L121
Fig.2a of your paper shows that the top MSRF-subnet output (n13 in your code) is used in the shape stream. However, this is currently not implemented in your code, only...