Technology Category
Computer Vision: Motion & TrackingMachine Learning: Learning on the Edge & Model CompressionIntelligent Robots: Multimodal Perception & Sensor Fusion
EDCFlow: Exploring Temporally Dense Difference Maps for Event-based Optical Flow Estimation
🤖 AI Summary
Existing event-camera optical flow methods suffer from high redundancy in cost volume computation and poor scalability to high-resolution inputs. To address these issues, this paper proposes EDCFlow, a lightweight network. Methodologically: (i) it introduces the first attention-driven multi-scale temporal feature differencing layer to explicitly model dynamic changes in event streams; (ii) it designs an adaptive fusion mechanism for high- and low-resolution motion features, balancing detail fidelity and computational efficiency; and (iii) it supports plug-and-play integration of RAFT-style refinement modules. Evaluated on multiple benchmarks, EDCFlow achieves state-of-the-art (SOTA) performance with significantly reduced computational complexity—averaging 38% lower than prior SOTA methods—while demonstrating strong generalization capability. Notably, it substantially improves structural fidelity and edge-detail accuracy in high-resolution flow fields.
Technology Category
Application Category
📝 Abstract
Recent learning-based methods for event-based optical flow estimation utilize cost volumes for pixel matching but suffer from redundant computations and limited scalability to higher resolutions for flow refinement. In this work, we take advantage of the complementarity between temporally dense feature differences of adjacent event frames and cost volume and present a lightweight event-based optical flow network (EDCFlow) to achieve high-quality flow estimation at a higher resolution. Specifically, an attention-based multi-scale temporal feature difference layer is developed to capture diverse motion patterns at high resolution in a computation-efficient manner. An adaptive fusion of high-resolution difference motion features and low-resolution correlation motion features is performed to enhance motion representation and model generalization. Notably, EDCFlow can serve as a plug-and-play refinement module for RAFT-like event-based methods to enhance flow details. Extensive experiments demonstrate that EDCFlow achieves better performance with lower complexity compared to existing methods, offering superior generalization.
Problem
Research questions and friction points this paper is trying to address.
Reduces redundant computations in event-based optical flow
Improves scalability to higher resolutions for flow refinement
Enhances motion representation with adaptive feature fusion
Innovation
Methods, ideas, or system contributions that make the work stand out.
Attention-based multi-scale temporal feature difference layer
Adaptive fusion of high and low resolution features
Plug-and-play refinement module for RAFT-like methods
🔎 Similar Papers
No similar papers found.
