PyTorch code and models for the DINOv2 self-supervised learning method.
 
 
Go to file
Etienne Guevel be41e7b2d3 Add env for cpu 2024-05-29 13:50:56 +02:00
.github/workflows Minor config tweaks (#246) 2023-09-30 20:01:59 +02:00
dinov2 fix stuff 2024-05-26 01:07:49 +02:00
notebooks Add env for cpu 2024-05-29 13:50:56 +02:00
notes initial changes 2024-05-14 15:11:51 +02:00
scripts Initial commit 2023-04-17 17:55:51 +00:00
tests initial changes 2024-05-14 15:11:51 +02:00
.gitignore Add depth estimation code (#184) 2023-08-31 14:57:50 +02:00
CODE_OF_CONDUCT.md Initial commit 2023-04-17 17:55:51 +00:00
CONTRIBUTING.md Initial commit 2023-04-17 17:55:51 +00:00
LICENSE Update license everywhere (#182) 2023-08-31 14:41:52 +02:00
MODEL_CARD.md Add new backbones trained with registers (#282) 2023-10-27 15:15:10 +02:00
README.md Add env for cpu 2024-05-29 13:50:56 +02:00
conda.yaml Add env for cpu 2024-05-29 13:50:56 +02:00
conda_cpu.yaml Add env for cpu 2024-05-29 13:50:56 +02:00
conda_or.yaml Add env for cpu 2024-05-29 13:50:56 +02:00
condas_or-extras.yaml Add env for cpu 2024-05-29 13:50:56 +02:00
hubconf.py Add new backbones trained with registers (#282) 2023-10-27 15:15:10 +02:00
pyproject.toml Initial commit 2023-04-17 17:55:51 +00:00
requirements-dev.txt Initial commit 2023-04-17 17:55:51 +00:00
requirements-extras.txt Add (optional) extras dependencies (#183) 2023-08-31 14:53:28 +02:00
requirements.txt Add env for cpu 2024-05-29 13:50:56 +02:00
setup.cfg Minor config tweaks (#246) 2023-09-30 20:01:59 +02:00
setup.py Add (optional) extras dependencies (#183) 2023-08-31 14:53:28 +02:00

README.md

DINOv2 for cell classification

This project is a fork of the DINOv2 repo published by meta, aiming to use the methodology presented in their papers to train a series of model on blood white cells images.
Developping a foundation model for blood white cells is interesting for several reasons:

  • The categories of blood white cells are not unanimous, and hematologists / datasets make different classes.
  • Some blood white cells present mutations that are visible on images, and those distinguishible features could be embedded by the model

Installing

Training

Most of the code used for the training