Research map / Video / Video segmentation and tracking
Efficient Meshflow and Optical Flow Estimation from Event Cameras
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces the HREM event-based meshflow dataset and the lightweight EEMFlow network for meshflow estimation, upgraded to EEMFlow+ with a confidence-induced detail completion module for optical flow.
Problem
Estimating spatially smooth sparse meshflow (and dense optical flow) from event camera data at high resolution, where existing event flow networks are slow and lack meshflow labels and high-resolution dynamic datasets.
Approach
A lightweight encoder-decoder with dilated feature correlation and group shuffle convolutions outputs full-resolution meshflow directly, and a confidence-guided CDC upsampling module sharpens motion boundaries when adapted to optical flow.
Key results
- EEMFlow reaches average EPE 5.50 on HREM (dt=1) vs 6.08 for FlowFormer, with 1.24M parameters and 7 ms inference (142.9 FPS).
- EEMFlow+ achieves 39.2 FPS on DSEC with 2PE 3.93, 3PE 2.15 and AE 2.67, state-of-the-art on several metrics.
- EEMFlow reduces parameters by 76% and inference time by 92% relative to ERAFT.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the self-corrector branch to CDC (ablation) | EEMFlow+ · DSEC | EPE: 0.89 -> 0.81 | Helped |
| add the self-correlation branch to CDC (ablation) | EEMFlow+ · DSEC | EPE: 0.89 -> 0.79 | Helped |
| add both the self-corrector and self-correlation branches to CDC (ablation) | EEMFlow+ · DSEC | EPE: 0.89 -> 0.75 | Helped |
Surprising findings
- Image-based FlowFormer struggles on difficult high-dynamic-range and fast-motion images where event-based methods remain superior.
Limitations
- Evaluation of meshflow relies on synthetic rendered data; real-world meshflow ground truth is not available.
- Future scene flow extension is not yet explored.
Experimental setup
Datasets: HREM DSEC
Benchmarks: HREM DSEC
Metrics: EPE 1PE 2PE 3PE AE FPS parameters
Baselines: EV-FlowNet PWCNet ERAFT SKFlow GMA KPAFlow FlowFormer TMA ADMFlow EFlowformer EVA-Flow
Builds on: ERAFT RAFT EV-FlowNet PWCNet
Closest papers by meaning
- EDCFlow: Exploring Temporally Dense Difference Maps for Event-based Optical Flow Estimation (CVPR 2025 · similarity 0.89)
- MemFlow: Optical Flow Estimation and Prediction with Memory (CVPR 2024 · similarity 0.87)
- V2V: Scaling Event-Based Vision through Efficient Video-to-Voxel Simulation (NeurIPS 2025 · similarity 0.85)
- E-MoFlow: Learning Egomotion and Optical Flow from Event Data via Implicit Regularization (NeurIPS 2025 · similarity 0.85)
- ZeroFlow: Scalable Scene Flow via Distillation (ICLR 2024 · similarity 0.85)
- Graph Neural Network Combining Event Stream and Periodic Aggregation for Low-Latency Event-based Vision (CVPR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.