microsoft / microsoft/XPretrain

Details of zero-shot performance on SSv2

Open
#44 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
511
Forks
35
PR merge metrics
No merged PRs in 30d

Description

Dear authors,

Great work!

I was wondering if you have the script to reproduce zero-shot numbers on SSv2 (Table 7).

Based on my experiments and also other papers [1, 2], I get 2.7% accuracy on the 174 classes in SSv2 with a frozen CLIP with mean pooling on per-frame features. Could you please elaborate on this discrepancy or what I may be missing?

[1] Videoprompter: an ensemble of foundational models for zero-shot video understanding. https://arxiv.org/pdf/2310.15324
[2] GPT4Vis: What Can GPT-4 Do for Zero-shot Visual Recognition? https://arxiv.org/pdf/2311.15732

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the SSv2 zero-shot setup underlying Table 7 and compare it with frozen CLIP mean pooling over per-frame features. Establish whether a reproduction script or configuration is available; done means a documented explanation of the 2.7% discrepancy or reproducible Table 7 numbers.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
computer-vision, machine-learning
Issue type
Documentation
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.