AI Paper Atlas

Research map / Detection and segmentation / Weakly supervised detection and segmentation

Weakly Supervised Class-Agnostic Motion Prediction for Autonomous Driving

Ruibo Li, Hanyu Shi, Ziang Fu, Zhe Wang, Guosheng Lin

CVPR 2023 · Detection and segmentation › Weakly supervised detection and segmentation · approach: Weakly supervised object localization · 14 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

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
Use fully or partially (1%, 0.1%) annotated FG/BG masks as weak supervision instead of motion labelsWeakMotionNet · nuScenes testMean error (static, slow, fast): ~35% better than self-supervised PillarMotion on all metricsImproved
Weakly supervised FG/BG masks (1%) vs. self-supervised PillarMotionFast speed group (>5m/s) · nuScenes testMean error: 1.7823 vs 3.5504Improved
Chamfer distance replaced by L1-norm distance in self-supervised loss (ablation)WeakMotionNet · nuScenes valMean error (static / slow / fast): static -42%, slow -37%, fast -11%Improved
Add past frame points to the target data (multi-frame Chamfer-L1) (ablation)WeakMotionNet · nuScenes valMean error speed >5m/s: 2.1229 to 1.7436Improved
Add confidence reweighting via multi-frame consistency (Consistency-aware Chamfer) (ablation)WeakMotionNet · nuScenes valMean error (static / slow / fast): static 0.1469 vs 0.2677 (improves); slow 0.4390 vs 0.5240 (improves); fast 1.7729 vs 1.7436 (slightly worse)Mixed
Add auxiliary FG/BG segmentation head with motion zeroed on predicted background (ablation)WeakMotionNet with CCD loss · nuScenes valMean error (static / slow / fast): static 0.1469 to 0.0558 (improves); slow 0.4390 to 0.4337 (roughly flat); fast 1.7729 to 1.7823 (roughly flat)Mixed
Remove Stage1 PreSegNet and generate FG/BG masks online from WeakMotionNet's segmentation head (ablation)two-stage framework · nuScenes valMean error (static / slow / fast): fast 8.9025 vs 1.7823; static 1.1976 vs 0.0558Hurt
Compute CCD loss on 2D BEV maps instead of mapping predictions to 3D points (ablation)WeakMotionNet · nuScenes valMean error speed >5m/s: 2.8176 vs 1.7823Hurt
Weakly supervised training on Waymo with 1% FG/BG masks vs. fully supervised MotionNetWeakMotionNet · WaymoMean error speed >5m/s: 1.5655 vs 0.9493 (worse)Hurt
Weakly supervised 100% masks vs. fully supervised MotionNet on Waymo static groupWeakMotionNet · WaymoMean error static: 0.0219 vs 0.0263Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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