AI Paper Atlas

Research map / Video / Video segmentation and tracking

TMA: Temporal Motion Aggregation for Event-based Optical Flow

Haotian Liu, Guang Chen, Sanqing Qu, Yanping Zhang, Zhijun Li, Alois Knoll, Changjun Jiang

ICCV 2023 · Video › Video segmentation and tracking · approach: Event-based video reconstruction · 54 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add temporal motion aggregation (TMA) in place of E-RAFTRAFT-style event flow · DSEC-Flow testEPE: 0.79 -> 0.74Improved
TMA needs fewer refinement iterations, reducing inference time versus E-RAFTTMA · DSEC-Flow testinference time (ms): 52 -> 30Improved
increase event splits from 1 to 5 (ablation)TMA · DSEC-Flow testEPE: 0.79 -> 0.74Improved
increase event splits from 5 to 7 (ablation)TMA · DSEC-Flow testEPE / time: EPE 0.74 -> 0.74, time 30 -> 50 msNo clear effect
add linear lookup versus no lookup (ablation)TMA · DSEC-Flow testEPE: 0.78 -> 0.74Improved
linear lookup instead of same sampling coordinates (ablation)TMA · DSEC-Flow testEPE: 0.76 -> 0.74Improved
add motion pattern aggregation layer (1 layer vs 0) (ablation)TMA · DSEC-Flow test1PE: 11.36 -> 10.86Improved
pair intermediate features with last feature instead of circular pairing (ablation)Pattern aggregation module · DSEC-Flow testEPE: 0.76 -> 0.74Improved
stack 2 aggregation layers instead of 1 (ablation)TMA · DSEC-Flow testEPE: 0.74 -> 0.75Hurt
increase refinement iterations from 2 to 6 (ablation)TMA · DSEC-Flow testEPE: 0.83 -> 0.74Improved
increase lookup search radius (ablation)TMA · DSEC-Flow testEPE: 0.74 (radius 3) vs 0.75 (radius 4)No clear effect
add TMA versus E-RAFT on MVSEC outdoor_day1 dt=4TMA · MVSEC outdoor_day1 (dt=4)EPE: 0.72 -> 0.70Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy