ShirAmir/dino-vit-features
Official implementation extracting deep ViT features from DINO as dense visual descriptors for co-segmentation and semantic correspondence tasks.
Not currently ranked — collecting fresh signals.
star history
The repository provides a PyTorch implementation for using pre-trained DINO Vision Transformer features as dense patch descriptors. It extracts features from a self-supervised ViT model and applies them to real-world vision tasks including co-segmentation, part segmentation, and point correspondence. The approach uses lightweight methods like clustering or binning on deep ViT features rather than task-specific learned components.
Frequently asked
- What is ShirAmir/dino-vit-features?
- Official implementation extracting deep ViT features from DINO as dense visual descriptors for co-segmentation and semantic correspondence tasks.
- Is dino-vit-features open source?
- Yes — ShirAmir/dino-vit-features is open source, released under the MIT license.
- What language is dino-vit-features written in?
- ShirAmir/dino-vit-features is primarily written in Python.
- How popular is dino-vit-features?
- ShirAmir/dino-vit-features has 468 stars on GitHub.
- Where can I find dino-vit-features?
- ShirAmir/dino-vit-features is on GitHub at https://github.com/ShirAmir/dino-vit-features.