Cascade R-CNN: Delving into High Quality Object Detection

CVPR 2018 Zhaowei CaiNuno Vasconcelos

In object detection, an intersection over union (IoU) threshold is required to define positives and negatives. An object detector, trained with low IoU threshold, e.g. 0.5, usually produces noisy detections... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT LEADERBOARD
Object Detection COCO minival Cascade R-CNN (ResNet-50-FPN+) box AP 40.3 # 41
AP50 59.4 # 29
AP75 43.7 # 28
APS 22.9 # 31
APM 43.7 # 28
APL 54.1 # 25
Object Detection COCO minival Cascade R-CNN (ResNet-101-FPN+, cascade) box AP 42.7 # 29
AP50 61.6 # 21
AP75 46.6 # 16
APS 23.8 # 27
APM 46.2 # 19
APL 57.4 # 16
Object Detection COCO test-dev Cascade R-CNN (ResNet-101-FPN+) box AP 38.8 # 61
AP50 61.1 # 47
AP75 41.9 # 58
APS 21.3 # 61
APM 41.8 # 56
APL 49.8 # 61
Object Detection COCO test-dev Cascade R-CNN (ResNet-50-FPN+) box AP 36.5 # 72
AP50 59 # 58
AP75 39.2 # 64
APS 20.3 # 65
APM 38.8 # 62
APL 46.4 # 63
Object Detection COCO test-dev Cascade R-CNN (ResNet-101-FPN+, cascade) box AP 42.8 # 38
AP50 62.1 # 43
AP75 46.3 # 38
APS 23.7 # 48
APM 45.5 # 40
APL 55.2 # 35
Object Detection COCO test-dev Cascade R-CNN (ResNet-50-FPN+, cascade) box AP 40.6 # 50
AP50 59.9 # 52
AP75 44 # 50
APS 22.6 # 53
APM 42.7 # 51
APL 52.1 # 49