In this assignment, we designed a self-supervised framework, based on the MoCo-V1 [1] and MoCo-V2 [2] papers. We started with an unsupervised training on the Imagenette dataset. Then, we used the extracted features from the unsupervised training step and trained a classifier in a fully supervised manner, and compared to an ImageNet pre-trained feature extractor.
You can create the project's environment from the env/env.yml file by running:
> cd env
> conda env create -f env.yml
> pip install -r main.txt
- Set parameters for training in
params.py - Train MoCo on Imagenette in
train_moco.py - Apply linear evaluation on Imagenette in
train_linear_classifier.py
[1] Momentum Contrast for Unsupervised Visual Representation Learning, He et al., 2020
[2] Improved Baselines with Momentum Contrastive Learning, Chen et al., 2020