MMDetection: Open MMLab Detection Toolbox and Benchmark

By Kai Chen, Jiangmiao Pang, Yuhang Cao, Yu Xiong, Xiaoxiao Li, Shuyang Sun, Wansen Feng, Ziwei Liu, Jiarui Xu, Zheng Zhang and 15 more. 2019

3,471 citations recorded by the source corpus.

We present MMDetection, an object detection toolbox that contains a rich set of object detection and instance segmentation methods as well as related components and modules. The toolbox started from a codebase of MMDet team who won the detection track of COCO Challenge 2018. It gradually evolves into a unified platform that covers many popular detection methods and contemporary modules. It not only includes training and inference codes, but also provides weights for more than 200 network models. We believe this toolbox is by far the most complete detection toolbox. In this paper, we introduce the various features of this toolbox. In addition, we also conduct a benchmarking study on different methods, components, and their hyper-parameters. 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. Code and models are available at this https URL. The project is under active development and we will keep this document updated.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Evaluation & Benchmarks

Citations in the corpus

Swin Transformer: Hierarchical Vision Transformer using Shifted Windows (2021, methodology): “For the ablation study, we consider four typical object detection frameworks: Cascade Mask R-CNN [26, 6], ATSS [71], RepPoints v2 [12], and Sparse RCNN [52] in mmdetection [10].”
SOCRATES: Introducing Depth in Visual Wildlife Monitoring Using Stereo Vision (2022, methodology): “We implemented our instance segmentation pipeline using mmdetection [46] and largely kept the default hyperparameters of the mmdetection model implementations, including basic color, horizontal flip, and random cropping augmentations.”
Learning Open-World Object Proposals without Learning to Classify (2021, methodology): “For all methods, we use the standard official models available in MMDetection [11] and train with the same default 1× training and testing pipeline.”
K-Net: Towards Unified Image Segmentation (2021, methodology): “The reported FPS is an average FPS obtained in three runs, where each run measures the FPS of a model through 400 iterations [6, 58].”
Prototypical Contrastive Learning of Unsupervised Representations (2020, methodology): “For object detection, We use the R50-FPN backbone for the Faster R-CNN detector available in the MMdetection [47] codebase.”
DADA: Differentiable Automatic Data Augmentation (2020, positive): “CO [18] with the better ResNet-50 model. We adopt three mainstream detectors RetinaNet [17], Faster R-CNN [23] and Mask R-CNN [8] in our experiments. For the same detector, we use the same setting as [2] except that the ResNet-50 model is trained with or without DADA policy. From Table 5, the perfor”
Fast Neural Network Adaptation via Parameter Remapping and Architecture Search (2020, methodology): “ShufﬂeNetV2-20 (Chen et al., 2019b) RetinaNet 13.19M 132.76B 32.1 MobileNetV2 (Sandler et al., 2018) 11.49M 133.05B 32.8 DetNAS (Chen et al., 2019b) 13.41M 133.26B 33.3 FNA 11.73M 133.03B 33.9 MobileNetV2 (Sandler et al., 2018) SSDLite 4.3M 0.8B 22.1 Mnasnet-92 (Tan et al., 2018) 5.3M 1.0B 22.9 FNA ”
AtomNAS: Fine-Grained End-to-End Neural Architecture Search (2019, methodology): “We use the open-source code MMDetection (Chen et al., 2019a).”
FairNAS: Rethinking Evaluation Fairness of Weight Sharing Neural Architecture Search (2019, methodology): “We follow the same setting as [27] and exploit MMDetection toolbox [5] for training.”
Benchmarking Robustness in Object Detection: Autonomous Driving when Winter is Coming (2019, methodology): “We additionally provide results for Faster R-CNN and Mask R-CNN models with deformable convolutions [Dai et al., 2017, Zhu et al., 2018] in Appendix D. Models were evaluated using the mmdetection toolbox [Chen et al., 2019b]; all models were trained and tested with standard hyperparameters.”
Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detection (2020, methodology): “λ0 (typically 2 as default, similarly in [3]) and λ1 (practically 14 , averaged over four directions) are the balance weights for LQ and LD, respectively.”
Deep High-Resolution Representation Learning for Visual Recognition (2019, methodology): “The results are obtained from MMDetection [17]. tection performance on COCO val . under two anchor-based frameworks: Faster R-CNN [118] and Cascade R-CNN [12], and two recently-developed anchor-free frameworks: FCOS [136] and CenterNet [36].”
ResNeSt: Split-Attention Networks (2020, neutral): “9 Faster-RCNN+DCN [13] ResNet101 [7] 42.”
Transformer in Transformer (2021, background): “Backbone Params (M) Epochs AP AP50 AP75 APS APM APL ResNet-50 [21, 5] 41.”
End-to-End Semi-Supervised Object Detection with Soft Teacher (2021, methodology): “Our implementation and hyper-parameters are based on MMDetection [4].”
Swin Transformer V2: Scaling Up Capacity and Resolution (2021, methodology): “COCO object detection We use cascade mask RCNN [8,28] implemented in mmdetection [11] as the object detection framework.”
PVT v2: Improved baselines with Pyramid Vision Transformer (2021, methodology): “Following common practices [21, 12, 3], we adopt 1 × or 3 × training schedule ( i.e. , 12 or 36 epochs) to train all detection models.”