Research map / 3D vision / 3D generation
Point4Cast: Streaming Dynamic Scene Reconstruction and Forecasting
Xinhang Liu, Pedro Miraldo, Suhas Lohit, Huaizu Jiang, Naoko Sawada, Yu-Wing Tai, Chi-Keung Tang, Moitreya Chatterjee
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 |
|---|---|---|---|
| use learned time embedding instead of sinusoidal encoding (ablation) | Point4Cast (cross-attention) · PointOdyssey reconstruction | Accuracy: 0.470 -> 0.437 | Improved |
| use learned time embedding instead of sinusoidal encoding (ablation) | Point4Cast (cross-attention) · PointOdyssey reconstruction | Completion: 0.502 -> 0.492 | Improved |
| use FiLM conditioning instead of cross-attention for time (ablation) | Point4Cast · PointOdyssey reconstruction | Accuracy: 0.437 -> 0.428 | Improved |
| use FiLM conditioning instead of cross-attention for time (ablation) | Point4Cast · PointOdyssey reconstruction | Completion: 0.492 -> 0.472 | Improved |
| swap VGGT backbone for CUT3R backbone (ablation) | Point4Cast · PointOdyssey reconstruction | Accuracy: 0.428 -> 0.410 | Mixed |
| add Point4Cast streaming spacetime representation on CUT3R backbone | CUT3R · PointOdyssey reconstruction | Accuracy: 0.530 -> 0.410 | Improved |
| add Point4Cast streaming spacetime representation on VGGT backbone | StreamVGGT · TAPVid-3D reconstruction (zero-shot) | Accuracy: 0.817 -> 0.711 | Improved |
| replace frame generation or scene-flow continuation with intrinsic time-conditioned forecasting | VGGT · PointOdyssey forecasting next frame | Accuracy: 0.561 -> 0.481 | Improved |
| replace frame generation or scene-flow continuation with intrinsic forecasting | CUT3R scene-flow continuation · PointOdyssey forecasting next 10 frames | Accuracy: 0.716 -> 0.561 | Improved |
| use Point4Cast instead of external forecasting for 10-frame horizon | StreamVGGT frame generation · TAPVid-3D forecasting next 10 frames | Accuracy: 1.693 -> 1.259 | Improved |
| replace forecasting with MonST3R frame generation on 10 frames | Point4Cast CUT3R · TAPVid-3D forecasting next 10 frames | Completion: 1.326 -> 1.471 | Hurt |
| add spacetime-representation scene flow, no flow supervision | MonST3R · PointOdyssey scene flow estimation | EPE: 2.058 -> 1.355 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VGGT — VGGT: Visual Geometry Grounded Transformer (CVPR 2025) · better on 4 of 4 shared results
- MonST3R — MonST3R: A Simple Approach for Estimating Geometry in the Presence of Motion (ICLR 2025) · better on 3 of 3 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
- IGGT4D: Streaming 4D Instance-Grounded Geometry Transformer (NeurIPS 2026 · similarity 0.89)
- Flux4D: Flow-based Unsupervised 4D Reconstruction (NeurIPS 2025 · similarity 0.89)
- Stereo4D: Learning How Things Move in 3D from Internet Stereo Videos (CVPR 2025 · similarity 0.89)
- World from Motion: Generative Dynamic Gaussian Reconstruction from Monocular Video (NeurIPS 2026 · similarity 0.88)
- DreamScene4D: Dynamic Multi-Object Scene Generation from Monocular Videos (NeurIPS 2024 · similarity 0.88)
- 4D3R: Motion-Aware Neural Reconstruction and Rendering of Dynamic Scenes from Monocular Videos (NeurIPS 2025 · similarity 0.88)
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.
