SMCA-DETR
SMCA-DETR copied to clipboard
Fast convergence of detr with spatially modulated co-attention
Usage
There are no extra compiled components in SMCA DETR and package dependencies are minimal, so the code is very simple to use. We provide instructions how to install dependencies via conda. First, clone the repository locally:
git clone https://github.com/facebookresearch/detr.git
Then, install PyTorch 1.5+ and torchvision 0.6+:
conda install -c pytorch pytorch torchvision
Install pycocotools (for evaluation on COCO) and scipy (for training):
conda install cython scipy
pip install -U 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'
That's it, should be good to train and evaluate detection models.
(optional) to work with panoptic install panopticapi:
pip install git+https://github.com/cocodataset/panopticapi.git
Data preparation
Download and extract COCO 2017 train and val images with annotations from http://cocodataset.org. We expect the directory structure to be the following:
path/to/coco/
annotations/ # annotation json files
train2017/ # train images
val2017/ # val images
Training
To train Single Scale SMCA on a single node with 8 gpus for 300 epochs run:
python -m torch.distributed.launch --nproc_per_node=8 --use_env main.py --coco_path /path/to/coco --batch_size 2 --lr_drop 40 --num_queries 300 --epochs 50 --dynamic_scale type3 --output_dir smca_single_scale
A single epoch takes 30 minutes, so 50 epoch training takes around 25 hours on a single machine with 8 V100 cards.
Object Detection
Model Zoo
name | dataset | backbone | schedule | box AP | |
---|---|---|---|---|---|
0 | SMCA(single scale) | MSCOCO | R50 | 50 | 41.0 |
1 | SMCA-Container(single scale) | MSCOCO | Container-S-Light | 50 | 44.2 |
2 | SMCA-Container(single scale) | MSCOCO | Container-M | 50 | 47.3 |
3 | SMCA(single scale) | MSCOCO | R50 | 108 | 42.7 |
4 | SMCA(single scale) | MSCOCO | R50 | 250 | 43.5 |
5 | SMCA(multi scale) | MSCOCO | R50 | 50 | 43.7 |
6 | SMCA(New multi scale) | MSCOCO | R50 | 50 | 44.4 |
7 | SMCA | Visual Genome | R50 | 50 | coming soon |
Panoptic Segmentation
Model Zoo
name | dataset | backbone | schedule | PQ | SQ | RQ | |
---|---|---|---|---|---|---|---|
1 | MASK-Former(single scale) | MSCOCO | R50 | 500 | 46.5 | 80.4 | 56.8 |
2 | SMCA-MASK-Former(single scale) | MSCOCO | R50 | 50 | 46.0 | 80.4 | 56.0 |
Release Steps
- Single-scale SMCA
- Single-scale SMCA with Container-Small
- Single-scale SMCA with Container-Medium
- New Multi-scale SMCA (Newly added Multi_scale_SMCA.zip, 9th Sep)
- SMCA-DETR for Fast Convergence of Panoptic Segmentation
Citation
If you find this repository useful, please consider citing our work:
@article{gao2021fast,
title={Fast convergence of detr with spatially modulated co-attention},
author={Gao, Peng and Zheng, Minghang and Wang, Xiaogang and Dai, Jifeng and Li, Hongsheng},
journal={arXiv preprint arXiv:2101.07448},
year={2021}
}
@article{gao2021container,
title={Container: Context Aggregation Network},
author={Gao, Peng and Lu, Jiasen and Li, Hongsheng and Mottaghi, Roozbeh and Kembhavi, Aniruddha},
journal={arXiv preprint arXiv:2106.01401},
year={2021}
}
@article{zheng2020end,
title={End-to-end object detection with adaptive clustering transformer},
author={Zheng, Minghang and Gao, Peng and Wang, Xiaogang and Li, Hongsheng and Dong, Hao},
journal={arXiv preprint arXiv:2011.09315},
year={2020}
}
Contributor
Peng Gao, Qiu Han, Minghang Zeng
Acknowledege
The project are borrowed heavily from DETR. Partially motivated by Sparse RCNN.