search

wisdomikezogwo / quilt1m

[NeurIPS 2023 Oral] Quilt-1M: One Million Image-Text Pairs for Histopathology.
https://quilt1m.github.io/
MIT License
138 stars 8 forks source link

Downstream tasks setting #3

Closed cheliu-computation closed 1 year ago

cheliu-computation commented 1 year ago

First thansk for your impressive work on meidcal VLP coummunity!

From your paper, there are many downstream tasks in the benchmark to evlaute the VLP model, could you provide the pipeline or script to prepare the downstream dataset and evaluation?

Best Regards

wisdomikezogwo commented 1 year ago

Thanks for your comments!

Added a few pointers on how to curate the datasets and evaluate QuiltNet on them in /eval/README.md. With time I'll update the ReadMe to include some of the data links directly, however, all the links are easily searchable and all the datasets are easily obtainable, so it shouldn't be a hassle parsing them, if you do have any issues worth so ever please let us know.

Again thanks.

cheliu-computation commented 1 year ago

Thanks a lot!

Now the dataset preparation is built on the csv and video downloading, will you upload the whole image-text dataset to huggingface or other place with raw image format rather than video?

wisdomikezogwo commented 1 year ago

We belive that violates Youtube policies, and so we employ you to leverage open-source packages to download said videos and follow the steps outlined in the readme to parse those videos into images and save them -- i.e reconstruct Quilt!

Hope this is helpful, if you have any other questions, I'm happy to help.

cheliu-computation commented 1 year ago

Thanks a lot!