facebookresearch / facebookresearch/perception_models

How to find the best layer in downstream task?

Open
#37 1 comment 0 reactions 1 assignee Claimed by @dbolya View on GitHub
PE question
Dominant language
Jupyter Notebook
Stars
2.4k
Forks
162
PR merge metrics
No merged PRs in 30d

Description

In the paper `Perception Encoder: The best visual embeddings are not at the output of the network`, Figure 13 presents the language alignment performance of features extracted from different layers.

Image

For PElang-G, were 47 experiments (through 1M pretraining + 70M finetuning each) conducted to find out the best layer? What would be the recommended approach to find out the best layer when I am facing a new downstream task?

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.