Automated Melanoma Recognition in Dermoscopy Images via Very Deep Residual Networks

First of all, very impressive work.

When I was trying to finetune the resnet50 work to reproduce your results, I found that the network tends to overfit the training data very easily and I got very low AP on the test data.

So I'm wondering are there any tricks in the fintuning. e.g.

1. what's the augmented size of your training set?
2. Did you employ any form of early stropping or just trained directly to 10000 iterations.

Really appreciate your help, thanks.

Thanks for sharing!

On the link provided to download a trained network from onedrive, The 'fc1000' layer has 1000 outputs. Is this correct for the trained model?

layer { bottom: "pool5" top: "fc1000" name: "fc1000" type: "InnerProduct" inner_product_param { num_output: 1000 } }

I am attempting to run the model via python and having a little difficulty interpreting the output. Any help would be greatly appreciated.

Many Thanks, Pjvance

Thank you for your sharing!

You said in the paper, "To train the FCRN, we first crop an sub-image from every original dermoscopy image with ground truth by automatically figuring out the smallest rectangle containing the lesion region and enlarging its length and width by 1.1 -1.3 times in order to include more neighboring pixels for training."

For segmantation, before input to the FCRN, whether you will resize the cropped sub-image to a fixed size? Or, you just use them as input?

Another question, whether the image size in a batch is same for segmantation?

Thank you!

Hi, @yulequan

Thanks for sharing this repo and the published paper. It somehow inspires me. Meanwhile, is there any chance that you provide your trained models to download? I mean, the caffemodel files, which can help me reveal the reported metrics such as AP, AUC, etc.

Since the augmented data is too large, I think the trained models are small and is convenient to download. Would that be convenient, please?

Hi Yulequan,

Thanks for sharing the codes for your paper. Could you kindly share the augmented dataset used in the codes for replication? Many thanks in ahead

Hi @yulequan ,

I am trying to reproduce your segmentation results.

I want to understand what specifically you have in your input .list file. Do you have file paths like ISIC_0000000.jpg ISIC_0000000_Segmentation.png (after cropping with respect to segmentation mask, then resizing to 480x480) at each row of the file? Can you give an example of one row from your .list file?

thanks in advance.

Thanks a lot for sharing the implementation. However, i find difficulty in going through the files. You are using three models for ResNet. Each model has a Fully connected layer of 1000 output. Did you apply segmentation with classification in the same network?