Deformable ConvNets v2: More Deformable, Better Results

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.

References in corpus (8)

• Distilling the Knowledge in a Neural Network
• A simple neural network module for relational reasoning
• Visualizing Deep Neural Network Decisions: Prediction Difference Analysis
• Non-local Neural Networks
• Discovering objects and their relations from entangled scene representations
• Visual Interaction Networks
• Relation Networks for Object Detection
• Learning Region Features for Object Detection

Cited by in corpus (27)

• A Survey of Deep Learning-based Object Detection
• GCNet: Non-local Networks Meet Squeeze-Excitation Networks and Beyond
• DetNAS: Backbone Search for Object Detection
• CornerNet-Lite: Efficient Keypoint Based Object Detection
• Benchmarking Robustness in Object Detection: Autonomous Driving when Winter is Coming
• Deep Learning in Mobile and Wireless Networking: A Survey
• An Empirical Study of Spatial Attention Mechanisms in Deep Networks
• EDVR: Video Restoration with Enhanced Deformable Convolutional Networks
• Efficient Neural Architecture Transformation Searchin Channel-Level for Object Detection
• An Analysis of Pre-Training on Object Detection
• NAS-FCOS: Fast Neural Architecture Search for Object Detection
• Consistent Optimization for Single-Shot Object Detection
• Contrastive Learning for Compact Single Image Dehazing
• Delving Deep into Pixel Alignment Feature for Accurate Multi-view Human Mesh Recovery
• ICDAR 2019 Competition on Large-scale Street View Text with Partial Labeling -- RRC-LSVT
• Adaptively Connected Neural Networks
• Face Detection with Feature Pyramids and Landmarks
• Ground-aware Monocular 3D Object Detection for Autonomous Driving
• One Self-Configurable Model to Solve Many Abstract Visual Reasoning Problems
• 2nd Place Solution in Google AI Open Images Object Detection Track 2019
• Adaptive Context Encoding Module for Semantic Segmentation
• Quality-Aware Network for Human Parsing
• AABO: Adaptive Anchor Box Optimization for Object Detection via Bayesian Sub-sampling
• Global Context Networks
• PDWN: Pyramid Deformable Warping Network for Video Interpolation
• 2nd Place Solution to Instance Segmentation of IJCAI 3D AI Challenge 2020
• Recognition of Russian traffic signs in winter conditions. Solutions of the "Ice Vision" competition winners