sayakpaul/probing-vits

Probing the representations of Vision Transformers.

38
/ 100
Emerging

Provides TensorFlow implementations of ViT, DeiT, and DINO with multiple probing techniques including attention rollout, mean attention distance, positional embedding visualization, and per-head attention extraction. Pre-trained weights from official codebases are loaded and validated against ImageNet-1k benchmarks. Interactive Hugging Face Spaces demos enable real-time attention visualization on custom images, complemented by Jupyter notebooks demonstrating video-to-attention heatmap generation and representation analysis.

340 stars. No commits in the last 6 months.

Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 10 / 25
Maturity 16 / 25
Community 12 / 25

How are scores calculated?

Stars

340

Forks

22

Language

Jupyter Notebook

License

Apache-2.0

Last pushed

Oct 05, 2022

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/transformers/sayakpaul/probing-vits"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.