M2Det: A Single-Shot Object Detector based on Multi-Level Feature Pyramid Network

12 Nov 2018Qijie ZhaoTao ShengYongtao WangZhi TangYing ChenLing CaiHaibin Ling

Feature pyramids are widely exploited by both the state-of-the-art one-stage object detectors (e.g., DSSD, RetinaNet, RefineDet) and the two-stage object detectors (e.g., Mask R-CNN, DetNet) to alleviate the problem arising from scale variation across object instances. Although these object detectors with feature pyramids achieve encouraging results, they have some limitations due to that they only simply construct the feature pyramid according to the inherent multi-scale, pyramidal architecture of the backbones which are actually designed for object classification task... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Object Detection COCO minival M2Det (VGG-16, 320x320) box AP 33.2 # 62
AP50 52.2 # 46
APS 15 # 39
APM 38.2 # 40
APL 49.1 # 40
Object Detection COCO minival M2Det (ResNet-1o1, 320x320) box AP 34.1 # 60
AP50 53.7 # 44
APS 15.9 # 38
APM 39.5 # 37
APL 49.3 # 39
Object Detection COCO test-dev M2Det (ResNet-101, single-scale) box AP 38.8 # 64
AP50 59.4 # 60
AP75 41.7 # 68
APS 20.5 # 68
APM 43.9 # 53
APL 53.4 # 49
Object Detection COCO test-dev M2Det (VGG-16, single-scale) box AP 41.0 # 49
AP50 59.7 # 59
AP75 45 # 52
APS 22.1 # 61
APM 46.5 # 42
APL 53.8 # 48
Object Detection COCO test-dev M2Det (ResNet-101, multi-scale) box AP 43.9 # 35
AP50 64.4 # 33
AP75 48 # 38
APS 29.6 # 23
APM 49.6 # 23
APL 54.3 # 46
Object Detection COCO test-dev M2Det (VGG-16, multi-scale) box AP 44.2 # 34
AP50 64.6 # 31
AP75 49.3 # 32
APS 29.2 # 25
APM 47.9 # 31
APL 55.1 # 42

Methods used in the Paper