This is a PyTorch implementation of the PLRC paper:
@inproceedings{bai2022point,
title={Point-Level Region Contrast for Object Detection Pre-Training},
author={Bai, Yutong and Chen, Xinlei and Kirillov, Alexander and Yuille, Alan and Berg, Alexander C},
booktitle={CVPR},
year={2022}
}
Install PyTorch and ImageNet dataset following the official PyTorch ImageNet training code.
This implementation only supports multi-gpu, DistributedDataParallel training, which is faster and simpler; single-gpu or DataParallel training is not supported.
To do unsupervised pre-training of a ResNet-50 model on ImageNet in an 8-gpu machine, run:
python main_plrc.py \
--dist-url 'tcp://localhost:10001' --multiprocessing-distributed --world-size 1 --rank 0
This script uses all the default hyper-parameters as described in the PRLC paper.
Our pre-trained ResNet-50 model and finetuned checkpoints on object detection can be downloaded as following:
Pretrained Model | Epoch | |
---|---|---|
Res50 | download link | 100 |
Finetuned Model | AP | AP50 | AP75 | |
---|---|---|---|---|
Res50 | download link | 58.2 | 82.7 | 65.1 |
The APs on Pascal VOC is averaged over 5 times.
Same as MoCo for object detection transfer, please see moco/detection.
For model visualzation, we provide an google colab for better illustration.
This project is under the CC-BY-NC 4.0 license. See LICENSE for details.