cjrd / r2o

Region-to-Object (R2O)

PyTorch implementation of Refine and Represent: Region-to-Object Representation Learning.

Installation of Apex is required to enable DDP.

To log metrics to wandb switch to enable_wandb:True in train_imagenet_300.yaml

Requirements

python>=3.9
pytorch>=1.10.0
pyyaml
tensorboardx
scikit-learn
scikit-image
opencv-python
pillow-simd
wandb

This repo uses torch.distributed.launch for pretraining:

python -m torch.distributed.launch --nproc_per_node=4--nnodes=32 --node_rank=0 --master_addr="" --master_port=12345 r2o_main.py {CONFIG_FILENAME}

Dataset Structure

imagenet
├── images
│   ├── train
│   │   ├── n01440764
│   │   ├── ...
│   │   ├── n15075141
│   ├── val
│   │   ├── n01440764
│   │   ├── ...
│   │   ├── n15075141

Pretrained Weights

We release pretrained weights pretrained on ImageNet-1k for 300 epochs in torchvision format.

Download

The evaluation baselines are as follows

Metric	Value
PASCAL VOC mIoU	76.7
Cityscapes mIoU	76.6
MS COCO $\text{AP}^{\text{bb}}$	41.7
MS COCO $\text{AP}^{\text{mk}}$	38.3

Reproduce Results

We use MMSegmentation for PASCAL VOC and Cityscapes semantic segmentation. We use detectron2 for MS COCO object detection and instance segmentation. The corresponding config can be found in eval folder.

Acknowledgement

This repo is based on the BYOL implementation from Yao: https://github.com/yaox12/BYOL-PyTorch and K-Means implementation from Ali Hassani https://github.com/alihassanijr/TorchKMeans