COCO Object detection with UniFormer

We currenent release the code and models for:

• Mask R-CNN

• Cascade Mask R-CNN

Updates

01/18/2022

[Initial commits]:

1. Models with Mask R-CNN.

2. Models with Cascade Mask R-CNN.

Model Zoo

The followed models and logs can be downloaded on Google Drive: total_models, total_logs.

We also release the models on Baidu Cloud: total_models (5v6i), total_logs (wr74).

Note

• All the models are pretrained on ImageNet-1K without Token Labeling and Layer Scale. Reason can be found in issue #12.

Mask R-CNN

Backbone	Lr Schd	box mAP	mask mAP	#params	FLOPs	Model	Log	Shell
UniFormer-Sh14	1x	45.6	41.6	41M	269G	google	google	run.sh/config
UniFormer-Sh14	3x+MS	48.2	43.4	41M	269G	google	google	run.sh/config
UniFormer-Bh14	1x	47.4	43.1	69M	399G	google	google	run.sh/config
UniFormer-Bh14	3x+MS	50.3	44.8	69M	399G	google	google	run.sh/config

Cascade Mask R-CNN

Backbone	Lr Schd	box mAP	mask mAP	#params	FLOPs	Model	Log	Shell
UniFormer-Sh14	3x+MS	52.1	45.2	79M	747G	google	google	run.sh/config
UniFormer-Bh14	3x+MS	53.8	46.4	107M	878G	google	google	run.sh/config

Usage

Installation

Please refer to get_started for installation and dataset preparation.

Training

1. Download the pretrained models in our repository.

2. Simply run the training scripts in exp as followed:

   bash ./exp/mask_rcnn_1x_hybrid_small/run.sh

   Or you can train other models as follower:

   # single-gpu training
   python tools/train.py <CONFIG_FILE> --cfg-options model.backbone.pretrained_path=<PRETRAIN_MODEL> [other optional arguments]
   
   # multi-gpu training
   tools/dist_train.sh <CONFIG_FILE> <GPU_NUM> --cfg-options model.backbone.pretrained_path=<PRETRAIN_MODEL> [other optional arguments] 

[Note]:

• We use hybrid MHRA to reduce training cost and set the corresponding hyperparameters in the config.py:

  window: False, # whether use window MHRA
  hybrid: True, # whether use hybrid MHRA
  window_size: 14, # size of window (>=14)

• To avoid out of memory, we use torch.utils.checkpoint in the config.py:

  use_checkpoint=True, # whether use checkpoint
  checkpoint_num=[0, 0, 8, 0], # index for using checkpoint in every stage

Testing

# single-gpu testing
python tools/test.py <CONFIG_FILE> <DET_CHECKPOINT_FILE> --eval bbox segm

# multi-gpu testing
tools/dist_test.sh <CONFIG_FILE> <DET_CHECKPOINT_FILE> <GPU_NUM> --eval bbox segm

Acknowledgement

This repository is built based on mmdetection and Swin Transformer repository.