Research map / Point clouds / Point cloud classification
DynaTok: Token-Based 4D Reconstruction from Partial Point Clouds
Weirong Chen, Keisuke Tateno, Hidenobu Matsuki, Michael Niemeyer, Daniel Cremers, Federico Tombari
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add cross-frame global attention (joint spatiotemporal alignment) to the token encoder (ablation) | DynaTok · DT4D-A | Accuracy (Acc.): 0.276 -> 0.027 | Improved |
| add cross-frame global attention (ablation) | DynaTok · DT4D-A | Normal Consistency: 0.531 -> 0.877 | Improved |
| increase tokens per frame from 128 to 256 (ablation) | DynaTok · DT4D-A | Completeness: 0.032 -> 0.026 | Improved |
| increase tokens per frame from 256 to 512 (ablation) | DynaTok · DT4D-A | Accuracy / Completeness / NC: Acc 0.027 -> 0.026; Comp 0.026 -> 0.030; NC 0.877 -> 0.881 | Mixed |
| increase number of training views from 2 to 8 (ablation) | DynaTok · DT4D-A | Completeness: 0.035 -> 0.026 | Improved |
| use DynaTok temporal aggregation instead of frame-wise 3D latent baselines | Kubric foreground · Kubric, Foreground + Background | Completeness: 0.020 -> 0.014 (vs TripoSG) | Improved |
| use DynaTok instead of Shape2VecSet on accuracy | Kubric Foreground + Background · Kubric, Foreground + Background | Accuracy: 0.009 -> 0.010 | No clear effect |
| use DynaTok instead of Motion2VecSets | DT4D-A Unseen Motion · DT4D-A, Unseen Motion | Completeness: 0.060 -> 0.021 | Improved |
| use DynaTok instead of Motion2VecSets for efficiency | 16-frame scene inference · Model complexity | Params / time per scene: 423M -> 19M params; ~45s -> ~22s | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Motion2VecSets — Motion2VecSets: 4D Latent Vector Set Diffusion for Non-rigid Shape Reconstruction and Tracking (CVPR 2024) · better on 6 of 6 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
- RFNet-4D: Joint Object Reconstruction and Flow Estimation from 4D Point Clouds (ECCV 2022 · similarity 0.88)
- DiffPMAE: Diffusion Masked Autoencoders for Point Cloud Reconstruction (ECCV 2024 · similarity 0.87)
- Register Any Point: Scaling 3D Point Cloud Registration by Flow Matching (ECCV 2026 · similarity 0.87)
- X4D-SceneFormer: Enhanced Scene Understanding on 4D Point Cloud Videos through Cross-Modal Knowledge Transfer (AAAI 2024 · similarity 0.86)
- Point2Real: Bridging the Gap between Point Cloud and Realistic Image for Open-World 3D Recognition (AAAI 2024 · similarity 0.85)
- Rectified Point Flow: Generic Point Cloud Pose Estimation (NeurIPS 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.
