Clean applications - Githubissues

Understanding-Visual-Datasets / VisDiff

Official implementation of "Describing Differences in Image Sets with Natural Language" (CVPR 2024 Oral)

https://understanding-visual-datasets.github.io/VisDiff-website/

Apache License 2.0

97 stars 10 forks source link

Clean applications #1

Closed lisadunlap closed 9 months ago

lisadunlap commented 9 months ago

updated applications (sans modeldiff) so that each has a generate_csv.py file and all the datasets are stored in data_paths
confirm that LaMem, ImageNet VS ImageNetV2, ResNet failure, and Diffusion run
change visualize logging to be media instead of a table
added in DiffusionProposer (still need to write visualization for that)
added summary metrics of top 5 hypotheses and scores

Still need to (in order of priority):

update readme
decide on final structure for applications (should the configs go in configs folder or applications?)
update configs to replicate results exactly where possible
clean and link original wandb projects