Deformable ConvNets v2: More Deformable, Better Results

CVPR 2019 Xizhou ZhuHan HuStephen LinJifeng Dai

The superior performance of Deformable Convolutional Networks arises from its ability to adapt to the geometric variations of objects. Through an examination of its adaptive behavior, we observe that while the spatial support for its neural features conforms more closely than regular ConvNets to object structure, this support may nevertheless extend well beyond the region of interest, causing features to be influenced by irrelevant image content... (read more)

PDF Abstract

Evaluation Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK COMPARE
Object Detection COCO minival Faster R-CNN (ResNet-101, DCNv2) box AP 41.7 # 27
Object Detection COCO minival Faster R-CNN (ResNet-101, DCNv2) APS 22.2 # 26
Object Detection COCO minival Faster R-CNN (ResNet-101, DCNv2) APM 45.8 # 16
Object Detection COCO minival Faster R-CNN (ResNet-101, DCNv2) APL 58.7 # 6
Object Detection COCO minival Mask R-CNN (ResNet-101, DCNv2) box AP 43.1 # 19
Object Detection COCO test-dev DCNv2 (ResNet-101, multi-scale) box AP 46.0 # 15
Object Detection COCO test-dev DCNv2 (ResNet-101, multi-scale) AP50 67.9 # 4
Object Detection COCO test-dev DCNv2 (ResNet-101, multi-scale) AP75 50.8 # 11
Object Detection COCO test-dev DCNv2 (ResNet-101, multi-scale) APS 27.8 # 13
Object Detection COCO test-dev DCNv2 (ResNet-101, multi-scale) APM 49.1 # 10
Object Detection COCO test-dev DCNv2 (ResNet-101, multi-scale) APL 59.5 # 7