mirror of
https://github.com/openai/CLIP.git
synced 2026-11-05 01:37:26 +00:00
[GH-ISSUE #101] training on YFCC100M #92
Labels
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
openai/CLIP#92
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @gcouairon on GitHub (Apr 30, 2021).
Original GitHub issue: https://github.com/openai/CLIP/issues/101
First, thanks for the greak work !
In the paper, you report scores for training a RN50 model on the filtered subset of YFCC100M. (31.3% zero-shot ImageNet accuracy)
Did you use the same parameters as for training on the WIT dataset (in particular, the same batch size) ?
Also, did you train a VIT-B/32 network on that subset ? If yes, did it perform better than the RN50 model ?
Thank you !
@jongwook commented on GitHub (Jul 19, 2021):
We used mostly the same hyperparameters as for the full CLIP models, except a lower weight decay of 0.1 and smaller batch size of 2048. We didn't train a ViT model on YFCC100M subset; our intuition is that it'll perform worse due to the smaller dataset and shorter training.