learningtitans/isbi2017-part3

RECOD Titans participation at the ISBI 2017 challenge - Part 3

isbi2017-part3's Issues

Original tensorflow/models version unspecified in documentation

While the README does mention that this project is a (manual) fork of tensorflow/models, it does not mention the commit ID off of which it had been forked, and due to subsequent commits in tensorflow/models, links to resources within that repo are broken. Mentioning the commit hash in README would help readers alot.

A problem with ISIC Archive additional data

Hi, thanks for your great job!

I have downloaded all the Additional ISIC Archive Images in the csv file deploy2017.csv. But I realized there are a lot of similar images, and it didn't help my model. Did you use all of that images to train your model?

Here's a print screen of the folder contains all ISIC archive images in the csv file.

PyTorch Pretrained Model

I found this repo very helpful, with nice documentation!

Would it be possible to help me with converting the pretrained model to PyTorch?

the more detail about meta-learning svm?

I am reading your paper "RECOD Titans at ISIC Challenge 2017". You team have did so much experiments and think deeply. I am really impressed on you idea of meta-learning svm. Is It a new scheme of Ensemble learning? I am nor sure, I guess you concatenating different models' feature and train a new svm classification? I hope could you give me a more detail introduction or some material? thanks!

more than 3 channels input

Dear Eduardo Valle,
I want to use your model for a different task with some images which are not 3-channel images (more than 3). I am trying to make tf records of my data by changing the convert_skin_lesion.py file. Do you think it is possible?

Thanks

pickle files

I cannot find "classes_to_sizes.pkl" and "splits_to_sizes.pkl". Can you upload them or tell me what I should put in those files please?

Thank you for sharing your code :)

Optimizer configuration

Hi

Could you explain what the intuition behind the immense values for epsilon and decay in RMSProp is?

thx,
simon