search

laura-rieger / deep-explanation-penalization

Code for using CDEP from the paper "Interpretations are useful: penalizing explanations to align neural networks with prior knowledge" https://arxiv.org/abs/1909.13584
MIT License
127 stars 14 forks source link

Identifying Colorful Patches #1

Closed geoffreyangus closed 4 years ago

geoffreyangus commented 4 years ago

Hello all,

First of all, I want to say great job on this paper and thank you so much for making the code available in such a usable fashion. I am trying to reproduce some of your results for the ISIC dataset. Specifically, I am trying to generate labels for benign images that identify the colorful patches. Is this something that can be done using your codebase? Let me know, thanks!

laura-rieger commented 4 years ago

Hi Geoffrey, glad that you enjoy the work! Just to make sure, you want to obtain labels identifying images with colorful patches (i.e. image has patch -> 1; image has no patch ->0)?

geoffreyangus commented 4 years ago

Yes exactly!

On Mon, Jan 13, 2020 at 7:54 AM Laura Rieger notifications@github.com wrote:

Hi Geoffrey, glad that you enjoy the work! Just to make sure, you want to obtain labels identifying images with colorful patches (i.e. image has patch -> 1; image has no patch ->0)?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/laura-rieger/deep-explanation-penalization/issues/1?email_source=notifications&email_token=AHCXU35AVHIBSEURGWPYELTQ5SFEJA5CNFSM4KGENJY2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEIZHNTA#issuecomment-573732556, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHCXU3ZWCP54YWDULWI2UKLQ5SFEJANCNFSM4KGENJYQ .

laura-rieger commented 4 years ago

In the ISIC archive the images with patches all come from one study. I think the easiest way to get labels would be to download the ISIC images +metadata (script 00 and 01 in the ISIC folder) and filter for images from the SONIC study. You'll have to make a user on isic-archive.com and change password/ user name in config.json

geoffreyangus commented 4 years ago

Awesome. I have downloaded the ISIC images and metadata using these first two scripts. Here's an example of the meta.csv DataFrame:

image

It is possible that the study is left out of the metadata script, since it only pulls out the clinical data. Looking at the raw API response, it looks like the following:

image

Do you recall which field would contain information regarding the type of study? Thanks so much for your help on this.

laura-rieger commented 4 years ago

If you go to the gallery on the ISIC website and filter for the SONIC dataset, below the images there's a button "Download as ZIP". Click on that, choose "Download all metadata in ISIC archive" and you'll get a zipped folder with json files for all images in the SONIC study named with the ID of the respective images. You can use those filenames to create the labels.

geoffreyangus commented 4 years ago

understood, thank you so much!