Research map / 3D vision / 3D generation
Transformable Bottleneck Networks
Kyle Olszewski, Sergey Tulyakov, Oliver Woodford, Hao Li, Linjie Luo
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 segmentation supervision (ablation) | TBN novel view synthesis · ShapeNet chair NVS (reduced model) | SSIM: 0.920 -> 0.921 | No clear effect |
| add segmentation supervision (ablation) | TBN novel view synthesis · ShapeNet chair NVS (reduced model) | foreground L1: 0.182 -> 0.189 | No clear effect |
| synthesize 4 extra views at regular poses and aggregate their bottlenecks | TBN occupancy decoder · ShapeNet chair 3D reconstruction | IoU: .3042 -> .4507 | Improved |
| synthesize 4 extra views at regular poses and aggregate their bottlenecks | TBN occupancy decoder · ShapeNet car 3D reconstruction | IoU: .4664 -> .7128 | Improved |
| synthesize 4 extra views at regular poses and aggregate their bottlenecks | TBN occupancy decoder · ShapeNet aeroplane 3D reconstruction | IoU: .2699 -> .4661 | Improved |
| sample synthesized views at regular instead of random poses | TBN 3D reconstruction (+4 views) · ShapeNet chair 3D reconstruction | IoU: .3648 -> .4507 | Improved |
| sample synthesized views at regular instead of random poses | TBN 3D reconstruction (+4 views) · ShapeNet car 3D reconstruction | IoU: .5674 -> .7128 | Improved |
| use real random-pose views instead of synthesized random-pose views | TBN 3D reconstruction (+4 views) · ShapeNet chair 3D reconstruction | IoU: .3648 -> .3822 | Improved |
| add synthesized random-pose views | TBN 3D reconstruction · ShapeNet chair 3D reconstruction | IoU: .3042 -> .3648 | Improved |
| increase additional regular synthetic views from 8 to 9 | TBN 3D reconstruction · ShapeNet chair 3D reconstruction | IoU: .4630 -> .4561 | No clear effect |
| increase input views at test time from 1 to 4 | TBN NVS · ShapeNet car | L1: .091 -> .059 | Improved |
| increase input views at test time from 1 to 4 | TBN NVS · ShapeNet chair | SSIM: .895 -> .939 | Improved |
Closest papers by meaning
- Flow Guided Transformable Bottleneck Networks for Motion Retargeting (CVPR 2021 · similarity 0.87)
- 3D-Aware Scene Manipulation via Inverse Graphics (NeurIPS 2018 · similarity 0.86)
- RayNet: Learning Volumetric 3D Reconstruction With Ray Potentials (CVPR 2018 · similarity 0.85)
- Learning Compact 3D Representations from Feed-Forward Novel View Synthesis (CVPR 2026 · similarity 0.85)
- ZPressor: Bottleneck-Aware Compression for Scalable Feed-Forward 3DGS (NeurIPS 2025 · similarity 0.84)
- Show and Tell: Visually Explainable Deep Neural Nets via Spatially-Aware Concept Bottleneck Models (CVPR 2025 · similarity 0.84)
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.
