DisParQ operates on patch tokens from a frozen self-supervised vision transformer (DINOv2).
Discrete part concepts from frozen self-supervised backbones match base accuracy without labels
DisParQ learns a prototype dictionary assigning one concept per image patch, with quantized attributes per concept, and reconstructs backbone representations at almost no accuracy cost.
Research Lab
Adam Pardyl · Siddhartha Gairola · Sukrut Rao · Adam Wróbel · Bartosz Zieliński · Bernt Schiele · +1 more
Jagiellonian University · Max Planck Institute for Informatics · Ardigen SA
Research Digest··3 min read
The authors introduce DisParQ, a method that learns spatially grounded, discrete part concepts from a frozen DINOv2 backbone without class labels or language supervision.
Why this paper
From Max Planck Institute for Informatics and 2 others
In one line
DisParQ learns discrete, spatially grounded part concepts and attributes from frozen self-supervised vision models without labels or language, matching teacher accuracy.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§