Res2Net-detectron2
Res2Net-detectron2 copied to clipboard
Res2Net for Panoptic Segmentation based on detectron2 (SOTA results).
Res2Net for Panoptic Segmentation based on detectron2.
Introduction
We propose a novel building block for CNNs, namely Res2Net, by constructing hierarchical residual-like connections within one single residual block. The Res2Net represents multi-scale features at a granular level and increases the range of receptive fields for each network layer.
Performance on COCO dataset
Name | lr sched |
train mem (GB) |
box AP |
mask AP |
PQ | download |
---|---|---|---|---|---|---|
R50-FPN | 1x | 4.8 | 37.6 | 34.7 | 39.4 | model | metrics |
R50-FPN | 3x | 4.8 | 40.0 | 36.5 | 41.5 | model | metrics |
R101-FPN | 3x | 6.0 | 42.4 | 38.5 | 43.0 | model | metrics |
Res2Net101-FPN | 3x | 6.0 | 44.0 | 39.6 | 44.5 | model | metrics |
- Res2Net101 has the similar parameters and FLOPs compared with ResNet101.
- We only test and show results of Panoptic Segmentation on detectron2, the detection and instance segmentation on detectron2 is also supported in this repo.
- Res2Net ImageNet pretrained models are in Res2Net-PretrainedModels.
- More applications of Res2Net are in Res2Net-Github.
Usage
- Use the tools/convert-torchvision-to-d2.py to transfer the ImageNet pretrained model of Res2Net101 to detectron2 supported format. Or just download the converted model from this link.
- Use the command to train:
./tools/train_net.py --num-gpus 8 --config-file configs/COCO-PanopticSegmentation/panoptic_fpn_R2_101_3x.yaml
Citation
If you find this work or code is helpful in your research, please cite:
@article{gao2019res2net,
title={Res2Net: A New Multi-scale Backbone Architecture},
author={Gao, Shang-Hua and Cheng, Ming-Ming and Zhao, Kai and Zhang, Xin-Yu and Yang, Ming-Hsuan and Torr, Philip},
journal={IEEE TPAMI},
year={2020},
doi={10.1109/TPAMI.2019.2938758},
}
For more details of detectron2, please refer to the detectron2 repo.
Detectron2 is Facebook AI Research's next generation software system that implements state-of-the-art object detection algorithms. It is a ground-up rewrite of the previous version, Detectron, and it originates from maskrcnn-benchmark.
![](https://user-images.githubusercontent.com/1381301/66535560-d3422200-eace-11e9-9123-5535d469db19.png)
What's New
- It is powered by the PyTorch deep learning framework.
- Includes more features such as panoptic segmentation, densepose, Cascade R-CNN, rotated bounding boxes, etc.
- Can be used as a library to support different projects on top of it. We'll open source more research projects in this way.
- It trains much faster.
See our blog post to see more demos and learn about detectron2.
Installation
See INSTALL.md.
Quick Start
See GETTING_STARTED.md, or the Colab Notebook.
Learn more at our documentation. And see projects/ for some projects that are built on top of detectron2.
Model Zoo and Baselines
We provide a large set of baseline results and trained models available for download in the Detectron2 Model Zoo.
License
Detectron2 is released under the Apache 2.0 license.
Citing Detectron2
If you use Detectron2 in your research or wish to refer to the baseline results published in the Model Zoo, please use the following BibTeX entry.
@misc{wu2019detectron2,
author = {Yuxin Wu and Alexander Kirillov and Francisco Massa and
Wan-Yen Lo and Ross Girshick},
title = {Detectron2},
howpublished = {\url{https://github.com/facebookresearch/detectron2}},
year = {2019}
}