Research map / Video / Video generation
HouseTour: A Virtual Real Estate A(I)gent
Ata Çelen, Marc Pollefeys, Daniel Barath, Iro Armeni
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 residual diffusion over spline interpolation for trajectories | Catmull-Rom spline · HouseTour trajectory generation | Euclidean distance: 106.2 -> 73.9 | Improved |
| add residual diffusion over spline interpolation | Catmull-Rom spline · HouseTour trajectory generation | R@1m: 64.7% -> 69.4% | Improved |
| add residual diffusion over spline interpolation | Catmull-Rom spline · HouseTour trajectory generation | PSNR / SSIM: 14.22 -> 14.24 | No clear effect |
| sparser known poses (every 5th to 15th frame) (ablation) | Residual Diffuser · HouseTour trajectory generation | Euclidean distance: 41.5 -> 151.1 -> 371.2 | Hurt |
| add residual diffusion at every-10th-frame sampling (ablation) | Catmull-Rom spline · HouseTour, every 10th frame | Euclidean distance: 219.0 -> 151.1 | Improved |
| add residual diffusion at every-5th-frame sampling (ablation) | Catmull-Rom spline · HouseTour, every 5th frame | Euclidean distance: 56.0 -> 41.5 | Improved |
| add residual diffusion at every-15th-frame sampling (ablation) | Catmull-Rom spline · HouseTour, every 15th frame | Euclidean distance: 510.9 -> 371.2 | Improved |
| supervised fine-tuning on real estate summaries | Qwen2-VL-7B · HouseTour summarization | Bradley-Terry score: 0.04 -> 0.71 | Improved |
| add 3D trajectory features as a third modality | Qwen2-VL-7B (SFT) · HouseTour summarization | BLEU-4: 0.070 -> 0.090 | Improved |
| add 3D trajectory features as a third modality | Qwen2-VL-7B (SFT) · HouseTour summarization | Bradley-Terry score: 0.71 -> 0.79 | Improved |
| add 3D trajectory features as a third modality | Qwen2-VL-7B (SFT) · HouseTour summarization | CIDEr: 0.026 -> 0.021 | Hurt |
| provide 3D positions at test time (ablation) | Qwen2-VL-3D vs SFT · scenes with 3D positions | GPT-4o preference rate: 56% -> 67.5% | Improved |
Closest papers by meaning
- RoomTour3D: Geometry-Aware Video-Instruction Tuning for Embodied Navigation (CVPR 2025 · similarity 0.86)
- From Indoor to Open World: Revealing the Spatial Reasoning Gap in MLLMs (CVPR 2026 · similarity 0.84)
- Director3D: Real-world Camera Trajectory and 3D Scene Generation from Text (NeurIPS 2024 · similarity 0.84)
- SemCity: Semantic Scene Generation with Triplane Diffusion (CVPR 2024 · similarity 0.84)
- Learning from Videos for 3D World: Enhancing MLLMs with 3D Vision Geometry Priors (NeurIPS 2025 · similarity 0.84)
- SceneCraft: Layout-Guided 3D Scene Generation (NeurIPS 2024 · 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.
