https://github.com/dgcnz/repr
Code for "Relative Position and Scale Regression for Self-Supervised Pretraining"
https://github.com/dgcnz/repr
cv ssl vit
Last synced: about 2 months ago
JSON representation
Code for "Relative Position and Scale Regression for Self-Supervised Pretraining"
- Host: GitHub
- URL: https://github.com/dgcnz/repr
- Owner: dgcnz
- Created: 2024-11-12T08:23:11.000Z (11 months ago)
- Default Branch: main
- Last Pushed: 2025-07-26T20:10:38.000Z (2 months ago)
- Last Synced: 2025-08-07T20:26:34.190Z (2 months ago)
- Topics: cv, ssl, vit
- Language: Python
- Homepage:
- Size: 42.8 MB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# REPR
## Usage
Use `uv` package manager.
```
uv sync
```
Config files are at `fabric_configs/experiments`.
To run pretraining locally (to debug):
```
uv run python -m scripts.main_pretrain experiment=pretrain/in1k/partmae_v6_vit_b_16/4060ti
```
To run on snellius adjust the following script:
```
sbatch scripts/slurm/train_partmae_v6_h100.sh
```
Model definition is at:
```
src/models/components/partmae_v6.py
```
To run experiments, see (fabric_configs/experiments):
```
uv run python -m src.experiments.linear_classification.main_linear \
train.seed=0 \
data=style-imagenette \
model=partmaev6_b
```