PCL

Prototypical Contrastive Learning of Unsupervised Representations (Salesforce Research)

<img src="./img/PCL_framework.png" width="600">

This is a PyTorch implementation of the <a href="https://openreview.net/forum?id=KmykpuSrjcq">PCL paper</a>:

<pre> @inproceedings{PCL, title={Prototypical Contrastive Learning of Unsupervised Representations}, author={Junnan Li and Pan Zhou and Caiming Xiong and Steven C.H. Hoi}, booktitle={ICLR}, year={2021} }</pre>

Requirements:

• ImageNet dataset
• Python ≥ 3.6
• PyTorch ≥ 1.4
• <a href="https://github.com/facebookresearch/faiss">faiss-gpu</a>: pip install faiss-gpu
• pip install tqdm

Unsupervised Training:

This implementation only supports multi-gpu, DistributedDataParallel training, which is faster and simpler; single-gpu or DataParallel training is not supported.

To perform unsupervised training of a ResNet-50 model on ImageNet using a 4-gpu or 8-gpu machine, run:

<pre>python main_pcl.py \ -a resnet50 \ --lr 0.03 \ --batch-size 256 \ --temperature 0.2 \ --mlp --aug-plus --cos (only activated for PCL v2) \ --dist-url 'tcp://localhost:10001' --multiprocessing-distributed --world-size 1 --rank 0 \ --exp-dir experiment_pcl [Imagenet dataset folder] </pre>

Download Pre-trained Models

<a href="https://storage.googleapis.com/sfr-pcl-data-research/PCL_checkpoint/PCL_v1_epoch200.pth.tar">PCL v1</a>	<a href="https://storage.googleapis.com/sfr-pcl-data-research/PCL_checkpoint/PCL_v2_epoch200.pth.tar">PCL v2</a>

Linear SVM Evaluation on VOC

To train a linear SVM classifier on VOC dataset, using frozen representations from a pre-trained model, run:

<pre>python eval_svm_voc.py --pretrained [your pretrained model] \ -a resnet50 \ --low-shot (only for low-shot evaluation, otherwise the entire dataset is used) \ [VOC2007 dataset folder] </pre>

Linear SVM classification result on VOC, using ResNet-50 pretrained with PCL for 200 epochs:

Model	k=1	k=2	k=4	k=8	k=16	Full
PCL v1	46.9	56.4	62.8	70.2	74.3	82.3
PCL v2	47.9	59.6	66.2	74.5	78.3	85.4

k is the number of training samples per class.

Linear Classifier Evaluation on ImageNet

Requirement: pip install tensorboard_logger
To train a logistic regression classifier on ImageNet, using frozen representations from a pre-trained model, run:

<pre>python eval_cls_imagenet.py --pretrained [your pretrained model] \ -a resnet50 \ --lr 5 \ --batch-size 256 \ --id ImageNet_linear \ --dist-url 'tcp://localhost:10001' --multiprocessing-distributed --world-size 1 --rank 0 \ [Imagenet dataset folder] </pre>

Linear classification result on ImageNet, using ResNet-50 pretrained with PCL for 200 epochs:

PCL v1	PCL v2
61.5	67.6