motokimura / UniverseNet

Object detection. EfficientDet-D5 level COCO AP in 20 epochs. SOTA single-stage detector on Waymo Open Dataset.

UniverseNet

UniverseNets are state-of-the-art detectors for universal-scale object detection. Please refer to our paper for details. https://arxiv.org/abs/2103.14027

Changelog

• 22.06 (June 2022):
  • Add SwinV2, FocalNet, GBR COTS dataset
  • Update codes for mmdet 2.25.0, mmcv-full 1.4.4
• 21.12 (Dec. 2021):
  • Support finer scale-wise AP metrics
  • Add codes for TOOD, ConvMLP, PoolFormer
  • Update codes for PyTorch 1.9.0, mmdet 2.17.0, mmcv-full 1.3.13
• 21.09 (Sept. 2021):
  • Support gradient accumulation to simulate large batch size with few GPUs (example)
  • Add codes for CBNetV2, PVT, PVTv2, DDOD
  • Update and fix codes for mmdet 2.14.0, mmcv-full 1.3.9
• 21.04 (Apr. 2021):
  • Propose Universal-Scale object detection Benchmark (USB)
  • Add codes for Swin Transformer, GFLv2, RelationNet++ (BVR)
  • Update and fix codes for PyTorch 1.7.1, mmdet 2.11.0, mmcv-full 1.3.2
• 20.12 (Dec. 2020):
  • Add configs for Manga109-s dataset
  • Add ATSS-style TTA for SOTA accuracy (COCO test-dev AP 54.1)
  • Add UniverseNet 20.08s for realtime speed (> 30 fps)
• 20.10 (Oct. 2020):
  • Add variants of UniverseNet 20.08
  • Update and fix codes for PyTorch 1.6.0, mmdet 2.4.0, mmcv-full 1.1.2
• 20.08 (Aug. 2020): UniverseNet 20.08
  • Improve usage of batchnorm
  • Use DCN modestly by default for faster training and inference
• 20.07 (July 2020): UniverseNet+GFL
  • Add GFL to improve accuracy and speed
  • Provide stronger pre-trained model (backbone: Res2Net-101)
• 20.06 (June 2020): UniverseNet
  • Achieve SOTA single-stage detector on Waymo Open Dataset 2D detection
  • Win 1st place in NightOwls Detection Challenge 2020 all objects track

Features not in the original MMDetection

Methods and architectures:

• UniverseNets (arXiv 2021)
• Swin Transformer V2 (CVPR 2022)
• FocalNet (arXiv 2022)
• PoolFormer (CVPR 2022)
• ConvMLP (arXiv 2021)
• CBNetV2 (arXiv 2021)
• TOOD (ICCV 2021) supported
• PVTv2 (CVMJ 2022) stronger models
• PVT (ICCV 2021) supported
• Swin Transformer (ICCV 2021) stronger models
• DDOD (ACMMM 2021) supported
• GFLv2 (CVPR 2021)
• RelationNet++ (BVR) (NeurIPS 2020)
• SEPC (CVPR 2020)
• ATSS-style TTA (CVPR 2020)
• Test-time augmentation for ATSS and GFL merged

Benchmarks and datasets:

• USB (arXiv 2021)
• Waymo Open Dataset (CVPR 2020)
• Manga109-s dataset (MTAP 2017, IEEE MultiMedia 2020)
• NightOwls dataset (ACCV 2018)
• GBR COTS dataset (arXiv 2021)

Usage

Installation

See get_started.md.

Basic Usage

See MMDetection documents. Especially, see this document to evaluate and train existing models on COCO.

Examples

We show examples to evaluate and train UniverseNet-20.08 on COCO with 4 GPUs.

# evaluate pre-trained model
mkdir -p ${HOME}/data/checkpoints/
wget -P ${HOME}/data/checkpoints/ https://github.com/shinya7y/UniverseNet/releases/download/20.08/universenet50_2008_fp16_4x4_mstrain_480_960_2x_coco_20200815_epoch_24-81356447.pth
CONFIG_FILE=configs/universenet/universenet50_2008_fp16_4x4_mstrain_480_960_2x_coco.py
CHECKPOINT_FILE=${HOME}/data/checkpoints/universenet50_2008_fp16_4x4_mstrain_480_960_2x_coco_20200815_epoch_24-81356447.pth
GPU_NUM=4
bash tools/dist_test.sh ${CONFIG_FILE} ${CHECKPOINT_FILE} ${GPU_NUM} --eval bbox

# train model
CONFIG_FILE=configs/universenet/universenet50_2008_fp16_4x4_mstrain_480_960_2x_coco.py
CONFIG_NAME=$(basename ${CONFIG_FILE} .py)
WORK_DIR="${HOME}/logs/coco/${CONFIG_NAME}_`date +%Y%m%d_%H%M%S`"
GPU_NUM=4
bash tools/dist_train.sh ${CONFIG_FILE} ${GPU_NUM} --work-dir ${WORK_DIR} --seed 0

Even if you have one GPU, we recommend using tools/dist_train.sh and tools/dist_test.sh to avoid a SyncBN issue.

Citation

@article{USB_shinya_2021,
  title={{USB}: Universal-Scale Object Detection Benchmark},
  author={Shinya, Yosuke},
  journal={arXiv:2103.14027},
  year={2021}
}

License

Major parts of the code are released under the Apache 2.0 license. Plsease check NOTICE for exceptions.

Acknowledgements

Some codes are modified from the repositories of FocalNet, PoolFormer, ConvMLP, Swin Transformer, Swin Transformer Object Detection, RelationNet++, SEPC, PVT, CBNetV2, GFLv2, and NightOwls. When merging, please note that there are some minor differences from the above repositories and the original MMDetection repository.

 

OpenMMLab website ^HOT      OpenMMLab platform ^{TRY IT OUT}

 

📘Documentation | 🛠️Installation | 👀Model Zoo | 🆕Update News | 🚀Ongoing Projects | 🤔Reporting Issues

English | 简体中文

Introduction

MMDetection is an open source object detection toolbox based on PyTorch. It is a part of the OpenMMLab project.

The master branch works with PyTorch 1.5+.

Major features

• Modular Design

  We decompose the detection framework into different components and one can easily construct a customized object detection framework by combining different modules.

• Support of multiple frameworks out of box

  The toolbox directly supports popular and contemporary detection frameworks, e.g. Faster RCNN, Mask RCNN, RetinaNet, etc.

• High efficiency

  All basic bbox and mask operations run on GPUs. The training speed is faster than or comparable to other codebases, including Detectron2, maskrcnn-benchmark and SimpleDet.

• State of the art

  The toolbox stems from the codebase developed by the MMDet team, who won COCO Detection Challenge in 2018, and we keep pushing it forward.

Apart from MMDetection, we also released a library mmcv for computer vision research, which is heavily depended on by this toolbox.

What's New

2.25.0 was released in 1/6/2022:

• Support dedicated MMDetWandbHook hook
• Support ConvNeXt, DDOD, SOLOv2
• Support Mask2Former for instance segmentation
• Rename config files of Mask2Former

Please refer to changelog.md for details and release history.

For compatibility changes between different versions of MMDetection, please refer to compatibility.md.

Installation

Please refer to Installation for installation instructions.

Getting Started

Please see get_started.md for the basic usage of MMDetection. We provide colab tutorial and instance segmentation colab tutorial, and other tutorials for:

• with existing dataset
• with new dataset
• with existing dataset_new_model
• learn about configs
• customize_datasets
• customize data pipelines
• customize_models
• customize runtime settings
• customize_losses
• finetuning models
• export a model to ONNX
• export ONNX to TRT
• weight initialization
• how to xxx

Overview of Benchmark and Model Zoo

Results and models are available in the model zoo.

Architectures

Object Detection	Instance Segmentation	Panoptic Segmentation	Other
Fast R-CNN (ICCV'2015) Faster R-CNN (NeurIPS'2015) RPN (NeurIPS'2015) SSD (ECCV'2016) RetinaNet (ICCV'2017) Cascade R-CNN (CVPR'2018) YOLOv3 (ArXiv'2018) CornerNet (ECCV'2018) Grid R-CNN (CVPR'2019) Guided Anchoring (CVPR'2019) FSAF (CVPR'2019) CenterNet (CVPR'2019) Libra R-CNN (CVPR'2019) TridentNet (ICCV'2019) FCOS (ICCV'2019) RepPoints (ICCV'2019) FreeAnchor (NeurIPS'2019) CascadeRPN (NeurIPS'2019) Foveabox (TIP'2020) Double-Head R-CNN (CVPR'2020) ATSS (CVPR'2020) NAS-FCOS (CVPR'2020) CentripetalNet (CVPR'2020) AutoAssign (ArXiv'2020) Side-Aware Boundary Localization (ECCV'2020) Dynamic R-CNN (ECCV'2020) DETR (ECCV'2020) PAA (ECCV'2020) VarifocalNet (CVPR'2021) Sparse R-CNN (CVPR'2021) YOLOF (CVPR'2021) YOLOX (CVPR'2021) Deformable DETR (ICLR'2021) TOOD (ICCV'2021) DDOD (ACM MM'2021)	Mask R-CNN (ICCV'2017) Cascade Mask R-CNN (CVPR'2018) Mask Scoring R-CNN (CVPR'2019) Hybrid Task Cascade (CVPR'2019) YOLACT (ICCV'2019) InstaBoost (ICCV'2019) SOLO (ECCV'2020) PointRend (CVPR'2020) DetectoRS (ArXiv'2020) SOLOv2 (NeurIPS'2020) SCNet (AAAI'2021) QueryInst (ICCV'2021) Mask2Former (ArXiv'2021)	Panoptic FPN (CVPR'2019) MaskFormer (NeurIPS'2021) Mask2Former (ArXiv'2021)	Contrastive Learning SwAV (NeurIPS'2020) MoCo (CVPR'2020) MoCov2 (ArXiv'2020) Distillation Localization Distillation (CVPR'2022) Label Assignment Distillation (WACV'2022)

Components

Backbones	Necks	Loss	Common
VGG (ICLR'2015) ResNet (CVPR'2016) ResNeXt (CVPR'2017) MobileNetV2 (CVPR'2018) HRNet (CVPR'2019) Generalized Attention (ICCV'2019) GCNet (ICCVW'2019) Res2Net (TPAMI'2020) RegNet (CVPR'2020) ResNeSt (ArXiv'2020) PVT (ICCV'2021) Swin (CVPR'2021) PVTv2 (ArXiv'2021) ResNet strikes back (ArXiv'2021) EfficientNet (ArXiv'2021) ConvNeXt (CVPR'2022)	PAFPN (CVPR'2018) NAS-FPN (CVPR'2019) CARAFE (ICCV'2019) FPG (ArXiv'2020) GRoIE (ICPR'2020) DyHead (CVPR'2021)	GHM (AAAI'2019) Generalized Focal Loss (NeurIPS'2020) Seasaw Loss (CVPR'2021)	OHEM (CVPR'2016) Group Normalization (ECCV'2018) DCN (ICCV'2017) DCNv2 (CVPR'2019) Weight Standardization (ArXiv'2019) Prime Sample Attention (CVPR'2020) Strong Baselines (CVPR'2021) Resnet strikes back (ArXiv'2021)

Some other methods are also supported in projects using MMDetection.

FAQ

Please refer to FAQ for frequently asked questions.

Contributing

We appreciate all contributions to improve MMDetection. Ongoing projects can be found in out GitHub Projects. Welcome community users to participate in these projects. Please refer to CONTRIBUTING.md for the contributing guideline.

Acknowledgement

MMDetection is an open source project that is contributed by researchers and engineers from various colleges and companies. We appreciate all the contributors who implement their methods or add new features, as well as users who give valuable feedbacks. We wish that the toolbox and benchmark could serve the growing research community by providing a flexible toolkit to reimplement existing methods and develop their own new detectors.

Citation

If you use this toolbox or benchmark in your research, please cite this project.

@article{mmdetection,
  title   = {{MMDetection}: Open MMLab Detection Toolbox and Benchmark},
  author  = {Chen, Kai and Wang, Jiaqi and Pang, Jiangmiao and Cao, Yuhang and
             Xiong, Yu and Li, Xiaoxiao and Sun, Shuyang and Feng, Wansen and
             Liu, Ziwei and Xu, Jiarui and Zhang, Zheng and Cheng, Dazhi and
             Zhu, Chenchen and Cheng, Tianheng and Zhao, Qijie and Li, Buyu and
             Lu, Xin and Zhu, Rui and Wu, Yue and Dai, Jifeng and Wang, Jingdong
             and Shi, Jianping and Ouyang, Wanli and Loy, Chen Change and Lin, Dahua},
  journal= {arXiv preprint arXiv:1906.07155},
  year={2019}
}

License

This project is released under the Apache 2.0 license.

Projects in OpenMMLab

• MMCV: OpenMMLab foundational library for computer vision.
• MIM: MIM installs OpenMMLab packages.
• MMClassification: OpenMMLab image classification toolbox and benchmark.
• MMDetection: OpenMMLab detection toolbox and benchmark.
• MMDetection3D: OpenMMLab's next-generation platform for general 3D object detection.
• MMRotate: OpenMMLab rotated object detection toolbox and benchmark.
• MMSegmentation: OpenMMLab semantic segmentation toolbox and benchmark.
• MMOCR: OpenMMLab text detection, recognition, and understanding toolbox.
• MMPose: OpenMMLab pose estimation toolbox and benchmark.
• MMHuman3D: OpenMMLab 3D human parametric model toolbox and benchmark.
• MMSelfSup: OpenMMLab self-supervised learning toolbox and benchmark.
• MMRazor: OpenMMLab model compression toolbox and benchmark.
• MMFewShot: OpenMMLab fewshot learning toolbox and benchmark.
• MMAction2: OpenMMLab's next-generation action understanding toolbox and benchmark.
• MMTracking: OpenMMLab video perception toolbox and benchmark.
• MMFlow: OpenMMLab optical flow toolbox and benchmark.
• MMEditing: OpenMMLab image and video editing toolbox.
• MMGeneration: OpenMMLab image and video generative models toolbox.
• MMDeploy: OpenMMLab model deployment framework.