Research map / 3D vision / 3D generation
Rays as Pixels: Learning A Joint Distribution of Video and Camera Trajectories
Wonbong Jang, Shikun Liu, Soubhik Sanyal, Juan Camilo Perez, Kam Woh Ng, Sanskar Agrawal, Juan-Manuel Perez-Rua, Yiannis Douratsos, Tao Xiang
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 |
|---|---|---|---|
| represent cameras as raxels (VAE-compatible RGB maps) instead of Plucker embeddings (ablation) | Rays as Pixels · DL3DV-140 cycle self-consistency | FVD: 333.56 -> 68.17 | Improved |
| add decoupled self-cross attention (DSCA) (ablation) | Rays as Pixels · DL3DV-140 cycle self-consistency | FVD: 77.08 -> 68.17 | Improved |
| add cosine similarity loss (ablation) | Rays as Pixels · DL3DV-140 cycle self-consistency | translation error: 0.094 -> 0.018 | Improved |
| add cosine similarity loss (ablation) | Rays as Pixels · DL3DV-140 cycle self-consistency | FID: 9.48 -> 7.33 | Improved |
| increase diffusion steps for pose estimation from 2 to 20 (ablation) | Rays as Pixels · DL3DV-140 pose estimation | mRRA@30: 88.37 -> 85.90 | Hurt |
| use 1 diffusion step instead of 2 for pose estimation (ablation) | Rays as Pixels · DL3DV-140 pose estimation | mRRA@30: 88.37 -> 82.78 | Hurt |
| compare against VGGT pointmap-based pose estimator | Rays as Pixels · Re10K pose estimation | mRRA@30: 95.91 vs 98.07 | Hurt |
| use joint video-camera model for camera-controlled generation | Rays as Pixels · DL3DV-140 generation | FID: 17.74 (Wonderland) -> 9.73 | Improved |
| use joint model vs Kaleido on trajectory adherence | Rays as Pixels · DL3DV-140 generation | rotation error: 0.011 (Kaleido) vs 0.098 | Hurt |
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.
- VGGT — VGGT: Visual Geometry Grounded Transformer (CVPR 2025) · better on 0 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
- Cameras as Rays: Pose Estimation via Ray Diffusion (ICLR 2024 · similarity 0.92)
- ViVid-1-to-3: Novel View Synthesis with Video Diffusion Models (CVPR 2024 · similarity 0.88)
- DiffusionSfM: Predicting Structure and Motion via Ray Origin and Endpoint Diffusion (CVPR 2025 · similarity 0.88)
- Reconstruct, Inpaint, Test-Time Finetune: Dynamic Novel-view Synthesis from Monocular Videos (NeurIPS 2025 · similarity 0.87)
- AC3D: Analyzing and Improving 3D Camera Control in Video Diffusion Transformers (CVPR 2025 · similarity 0.87)
- Novel View Synthesis with Pixel-Space Diffusion Models (CVPR 2025 · similarity 0.87)
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.
