search

xwjabc / hed

A PyTorch reimplementation of Holistically-Nested Edge Detection
169 stars 42 forks source link

would be useful to use HED for binary edge detection task? #22

Open NdaAzr opened 4 years ago

NdaAzr commented 4 years ago

Thank you for sharing your implementation. I am wondering would be useful to use the HED network for the binary edge detection for the custom dataset?

Here is an example of my image and its ground truth:

Thank you in advance

xwjabc commented 4 years ago

I think it is possible. Since the image you provided seems to be a medical scanning, which might be different from the BSDS dataset, I would suggest you fine-tune the pre-trained HED model on your dataset first and then evaluate on your dataset.

NdaAzr commented 4 years ago

@xwjabc Thank you for your reply. How about training from scratch?

xwjabc commented 4 years ago

@xwjabc Thank you for your reply. How about training from scratch?

It should be also fine. However, sometimes the edge detection performance may not be very well when training from scratch since some medical imaging datasets are pretty small. In that case, pre-training on generic edge detection dataset and fine-tuning on the medical imaging datasets would be better.