[Feature] Support DDOD: Disentangle Your Dense Object Detector(ACM MM2021 oral)

[Irvingao]

Disentangle Your Dense Object Detector

Disentangle Your Dense Object Detector

Abstract

Deep learning-based dense object detectors have achieved great success in the past few years and have been applied to numerous multimedia applications such as video understanding. However, the current training pipeline for dense detectors is compromised to lots of conjunctions that may not hold. In this paper, we investigate three such important conjunctions: 1) only samples assigned as positive in classification head are used to train the regression head; 2) classification and regression share the same input feature and computational fields defined by the parallel head architecture; and 3) samples distributed in different feature pyramid layers are treated equally when computing the loss. We first carry out a series of pilot experiments to show disentangling such conjunctions can lead to persistent performance improvement. Then, based on these findings, we propose Disentangled Dense Object Detector(DDOD), in which simple and effective disentanglement mechanisms are designed and integrated into the current state-of-the-art dense object detectors. Extensive experiments on MS COCO benchmark show that our approach can lead to 2.0 mAP, 2.4 mAP and 2.2 mAP absolute improvements on RetinaNet, FCOS, and ATSS baselines with negligible extra overhead. Notably, our best model reaches 55.0 mAP on the COCO test-dev set and 93.5 AP on the hard subset of WIDER FACE, achieving new state-of-the-art performance on these two competitive benchmarks. Code is available at https://github.com/zehuichen123/DDOD.

Results and Models

Model	Backbone	Style	Lr schd	Mem (GB)	Inf time (fps)	box AP	Config	Download
DDOD-ATSS	ResNet50	pytorch	1x	#TODO	#TODO	41.6	config	model | log

Citation

@inproceedings{chen2021disentangle,
title={Disentangle Your Dense Object Detector},
author={Chen, Zehui and Yang, Chenhongyi and Li, Qiaofei and Zhao, Feng and Zha, Zheng-Jun and Wu, Feng},
booktitle={Proceedings of the 29th ACM International Conference on Multimedia},
pages={4939--4948},
year={2021}
}

Checklist

It's the second PR of DDOD from PR#7227 , which changes branch from master to dev. Other changes are as follow:

1. Rewrite README.md according to https://github.com/open-mmlab/mmdetection/blob/master/configs/retinanet/README.md#results-and-models.
2. Rename config file as ddod_r50_fp16_1x_coco.py.
3. Use _base _ to replace repeated code in the config file.
4. Add diou calculation function to bbox_overlaps.
5. add @torch.no_grad() to function.
6. Modify DDODHead docs.

---

Log

Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.416
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=1000 ] = 0.598
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=1000 ] = 0.452
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=1000 ] = 0.240
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=1000 ] = 0.450
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=1000 ] = 0.546
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.596
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=300 ] = 0.596
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=1000 ] = 0.596
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=1000 ] = 0.390
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=1000 ] = 0.636
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=1000 ] = 0.753

2022-02-26 22:24:22,587 - mmdet - INFO - Exp name: ddod_r50_1x_coco_fp16.py
2022-02-26 22:24:22,588 - mmdet - INFO - Epoch(val) [12][625] bbox_mAP: 0.4160, bbox_mAP_50: 0.5980, bbox_mAP_75: 0.4520, bbox_mAP_s: 0.2400, bbox_mAP_m: 0.4500, bbox_mAP_l: 0.5460, bbox_mAP_copypaste: 0.416 0.598 0.452 0.240 0.450 0.546

[chhluo]

Please run pip install pre-commit to fix lint problem first, refer to https://pre-commit.com/. Do we need src/mmtrack ?

[hhaAndroid]

@Irvingao You can resolve conflicts before proceeding with further development.

[ZwwWayne]

Kindly ping @Irvingao to resolve the conflict. Furthermore, will there be any progress in the following weeks?

[chhluo]

Please solve the conflicts and rebase on the latest dev branch