search

stegmuel / CrOC

This repo contains the code for the CVPR 2023 paper: "CrOC : Cross-View Online Clustering for Dense Visual Representation Learning".
https://openaccess.thecvf.com/content/CVPR2023/papers/Stegmuller_CrOC_Cross-View_Online_Clustering_for_Dense_Visual_Representation_Learning_CVPR_2023_paper.pdf
MIT License
21 stars 2 forks source link

Evaluation code #1

Open alexcbb opened 7 months ago

alexcbb commented 7 months ago

Hello,

First of all, thank you for your amazing work ! I was currently studying your method as it seemed to correlate with some ideas that I had. I am able to replicate the training (and it smoothly converge with the provided args) but it seems that no code is provided to replicate the evaluation pipeline. Did you plan to release this part or have you maybe some hindsights on how to replicate these evaluations ? Thank you in advance !

stegmuel commented 7 months ago

Hi,

Glad to hear that you were able to use our code. We did not plan on releasing the code for the evaluations.

For the unsupervised and linear segmentation evaluations, we used the code from MaskContrast (https://github.com/wvangansbeke/Unsupervised-Semantic-Segmentation) and leopart (https://github.com/MkuuWaUjinga/leopart).

For the video segmentation evaluation, we used the code from DINO (https://github.com/facebookresearch/dino) and SelfPatch (https://github.com/alinlab/SelfPatch).

Wangxq7 commented 4 months ago

@alexcbb Hello,

Have you solved the evaluation code? I have encountered some problems now. In addition, your homepage email does not show this information, could you please give me an email? I would like to ask you some questions, thank you in advance for your help!

alexcbb commented 4 months ago

@alexcbb Hello,

Have you solved the evaluation code? I have encountered some problems now. In addition, your homepage email does not show this information, could you please give me an email? I would like to ask you some questions, thank you in advance for your help!

Hello, unfortunately, my project requirements evolved and I didn't had the occasion to try out a way to evaluate the model.