Res2Net: A New Multi-scale Backbone Architecture

2 Apr 2019Shang-Hua GaoMing-Ming ChengKai ZhaoXin-Yu ZhangMing-Hsuan YangPhilip Torr

Representing features at multiple scales is of great importance for numerous vision tasks. Recent advances in backbone convolutional neural networks (CNNs) continually demonstrate stronger multi-scale representation ability, leading to consistent performance gains on a wide range of applications... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK USES EXTRA
TRAINING DATA
RESULT LEADERBOARD
Image Classification CIFAR-100 Res2NeXt-29 Percentage correct 83.44 # 16
Percentage error 16.56 # 10
Instance Segmentation COCO minival Faster R-CNN (Res2Net-50) mask AP 35.6 # 17
AP50 57.6 # 4
APL 53.7 # 3
APM 37.9 # 3
APS 15.7 # 3
Object Detection COCO minival Faster R-CNN (Res2Net-50) box AP 33.7 # 61
AP50 53.6 # 45
APS 14 # 40
APM 38.3 # 39
APL 51.1 # 34
Object Detection COCO minival Res2Net101+HTC box AP 47.5 # 9
AP50 66.5 # 10
AP75 51.3 # 8
APS 28.6 # 7
APM 51.6 # 6
APL 62.1 # 8
Instance Segmentation COCO minival Res2Net-101+HTC mask AP 41.3 # 6
Salient Object Detection DUT-OMRON DSS (Res2Net-50) MAE 0.071 # 3
F-measure 0.800 # 2
Salient Object Detection ECSSD DSS (Res2Net-50) MAE 0.056 # 3
F-measure 0.926 # 2
Salient Object Detection HKU-IS DSS (Res2Net-50) MAE 0.05 # 4
F-measure 0.905 # 2
Image Classification ImageNet Res2Net-50-299 Top 1 Accuracy 78.59% # 83
Top 5 Accuracy 94.12% # 62
Image Classification ImageNet Res2Net-101 Top 1 Accuracy 81.23% # 49
Top 5 Accuracy 94.43% # 55
Salient Object Detection PASCAL-S DSS (Res2Net-50) MAE 0.099 # 5
F-measure 0.841 # 2
Object Detection PASCAL VOC 2007 Faster R-CNN (Res2Net-50) MAP 74.4% # 19
Semantic Segmentation PASCAL VOC 2012 val Deeplab v3+ (Res2Net-101) mIoU 79.3% # 8

Methods used in the Paper