jk-the-ko / sar-cam Goto Github PK

Dear author, it's a great honor to see your work, would you mind making your dataset public？

Could you please describe the environment configuration and how to inference?

Hello! I would like to ask, would you mind updating the preprocessing code for the dataset? For example, use MATLAB to add speckle noise. There is also the "cropping of the image into a 64 × 64 patch size" mentioned in the paper. Is it a random cropping of the original image? Does this mean that the training and validation set image sizes are 64 × 64?
Looking forward to your reply.

Could you please send me the adjusted data set? I would be very grateful.

Hello, Author,
I successfully ran python train.py --project PROJECT_NAME --noisy-train-dir NOISY_IMAGE_TRAIN_DIR --clean-train-dir CLEAN_IMAGE_TRAIN_DIR --noisy-valid-dir NOISY_IMAGE_VALID_DIR --clean-valid-dir CLEAN_IMAGE_VALID_DIR and obtained the weight parameters SAR-CAM_best.pth. However, when I ran the second set of code for testing on the test set, I encountered a mismatch between the weights and the architecture when loading the weight parameters. The specific error code is:

RuntimeError: Error(s) in loading state_dict for SAR_CAM: Unexpected key(s) in state_dict: "conv_in.conv.bias", "conv_out.conv.bias", ... "up.up.1.conv_in.conv.bias", "up.bottleneck.conv.bias"...

I am not clear about the specific reason. I hope you can help to clarify. Thank you!

i got negative validation loss after training ? could you tell me how the model compare images. i have low and high imaes pair as a input images. i consider low as a noisy and high quality images as clean image.