Deformable ConvNets v2: More Deformable, Better Results

TL;DR

Deformable ConvNets v2 enhances the original model by improving focus on relevant image regions through increased modeling power and training strategies, leading to significant performance improvements in object detection and segmentation.

Contribution

This paper introduces a reformulated Deformable ConvNets with expanded deformation modeling and a feature mimicking training scheme for better object focus.

Findings

Significant performance gains over the original Deformable ConvNets

Leading results on COCO object detection and segmentation benchmarks

Enhanced ability to focus on pertinent image regions

Abstract

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. To address this problem, we present a reformulation of Deformable ConvNets that improves its ability to focus on pertinent image regions, through increased modeling power and stronger training. The modeling power is enhanced through a more comprehensive integration of deformable convolution within the network, and by introducing a modulation mechanism that expands the scope of deformation modeling. To effectively harness this enriched modeling capability, we guide network training via a proposed feature mimicking scheme that helps the network to learn features that reflect the object focus and classification power of R-CNN features. With the proposed contributions, this new version of Deformable ConvNets yields significant performance gains over the original model and produces leading results on the COCO benchmark for object detection and instance segmentation.