google-deepmind / google-deepmind/deepmind-research

Question about BYOL results

Open
#276 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Jupyter Notebook
Stars
15.2k
Forks
2.9k
PR merge metrics
No merged PRs in 30d

Description

Hello there!

I have been running some experiments with the [BYOL](https://github.com/deepmind/deepmind-research/tree/master/byol) framework. I used the fast iteration with imagenette since I was using a single GPU for the experiments.

Using `batch_size=256` and `ResNet50` , I ran the pretrain mode in the script and *apparently*, It achieved $100\%$ top 1 accuracy score.

- Is this normal?
- Where is this accuracy being computed in ImageNette, since it only has train and validation sets (no test set) ?

Thank u guys!

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.