DetectoRS: Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolution

CVPR 2021  ·  Siyuan Qiao, Liang-Chieh Chen, Alan Yuille ·

Many modern object detectors demonstrate outstanding performances by using the mechanism of looking and thinking twice. In this paper, we explore this mechanism in the backbone design for object detection. At the macro level, we propose Recursive Feature Pyramid, which incorporates extra feedback connections from Feature Pyramid Networks into the bottom-up backbone layers. At the micro level, we propose Switchable Atrous Convolution, which convolves the features with different atrous rates and gathers the results using switch functions. Combining them results in DetectoRS, which significantly improves the performances of object detection. On COCO test-dev, DetectoRS achieves state-of-the-art 55.7% box AP for object detection, 48.5% mask AP for instance segmentation, and 50.0% PQ for panoptic segmentation. The code is made publicly available.

PDF Abstract CVPR 2021 PDF CVPR 2021 Abstract

Datasets


Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Object Detection COCO test-dev DetectoRS (ResNeXt-101-64x4d, multi-scale) box mAP 55.7 # 44
AP50 74.2 # 9
AP75 61.1 # 11
APS 37.7 # 7
APM 58.4 # 11
APL 68.1 # 11
Object Detection COCO test-dev DetectoRS (ResNeXt-101-32x4d, multi-scale) box mAP 54.7 # 49
AP50 73.5 # 12
AP75 60.1 # 15
APS 37.4 # 9
APM 57.3 # 13
APL 66.4 # 17
Object Detection COCO test-dev DetectoRS (ResNeXt-101-32x4d, single-scale) box mAP 53.3 # 59
AP50 71.6 # 23
AP75 58.5 # 22
APS 33.9 # 25
APM 56.5 # 16
APL 66.9 # 13
Panoptic Segmentation COCO test-dev DetectoRS (ResNeXt-101-64x4d, multi-scale) PQ 50 # 15
PQst 37.2 # 19
PQth 58.5 # 10
Instance Segmentation COCO test-dev DetectoRS (ResNeXt-101-64x4d, multi-scale) mask AP 48.5 # 26
AP50 72.0 # 8
AP75 53.3 # 7
APS 31.6 # 6
APM 50.9 # 7
APL 61.5 # 9
Instance Segmentation COCO test-dev DetectoRS (ResNeXt-101-32x4d, multi-scale) mask AP 47.1 # 31
AP50 71.1 # 9
AP75 51.6 # 8
APS 30.3 # 8
APM 49.5 # 9
APL 59.6 # 11

Results from Other Papers


Task Dataset Model Metric Name Metric Value Rank Source Paper Compare
Object Detection AI-TOD DetectoRS (ResNet-50-FPN) AP 14.8 # 3
AP50 32.8 # 4
AP75 11.4 # 3
APvt 0.0 # 4
APt 10.8 # 4
APs 28.3 # 3
APm 28.0 # 3

Methods