Research map / Detection and segmentation / 3D detection and driving
SceneNet RGB-D: Can 5M Synthetic Images Beat Generic ImageNet Pre-Training on Indoor Segmentation?
John McCormac, Ankur Handa, Stefan Leutenegger, Andrew J. Davison
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 |
|---|---|---|---|
| pre-train on SceneNet RGB-D synthetic RGB instead of ImageNet VGG-16 init | U-Net RGB · NYUv2 (13 class, val) | mean IU: 33.7 -> 41.8 | Improved |
| pre-train on SceneNet RGB-D synthetic RGB instead of ImageNet VGG-16 init | U-Net RGB · NYUv2 (13 class, val) | class average accuracy: 47.6 -> 56.0 | Improved |
| pre-train on SceneNet RGB-D synthetic RGB instead of ImageNet VGG-16 init | U-Net RGB · NYUv2 (13 class, val) | pixel average accuracy: 60.2 -> 66.1 | Improved |
| pre-train on SceneNet RGB-D synthetic RGB instead of ImageNet VGG-16 init | U-Net RGB · SUN RGB-D (13 class, val) | mean IU: 40.3 -> 43.8 | Improved |
| pre-train on SceneNet RGB-D synthetic RGB instead of ImageNet VGG-16 init | U-Net RGB · SUN RGB-D (13 class, val) | class average accuracy: 56.8 -> 57.8 | Improved |
| pre-train on SceneNet RGB-D synthetic RGB instead of ImageNet VGG-16 init | U-Net RGB · SUN RGB-D (13 class, val) | pixel average accuracy: 71.2 -> 73.3 | Improved |
| add synthetic RGB pre-training vs no pre-training (ablation) | U-Net RGB · NYUv2 (13 class, val) | mean IU: 29.0 -> 41.8 | Improved |
| add synthetic RGB pre-training vs no pre-training (ablation) | U-Net RGB · SUN RGB-D (13 class, val) | mean IU: 36.5 -> 43.8 | Improved |
| add depth channel to synthetic pre-trained network (ablation) | U-Net pre-trained on SceneNet · NYUv2 (13 class, val) | mean IU: 41.8 -> 48.2 | Improved |
| add depth channel to synthetic pre-trained network (ablation) | U-Net pre-trained on SceneNet · SUN RGB-D (13 class, val) | mean IU: 43.8 -> 49.8 | Improved |
| add SceneNet pre-training to RGB-D network vs training from scratch (ablation) | U-Net RGB-D · NYUv2 (13 class, val) | mean IU: 36.9 -> 48.2 | Improved |
| add SceneNet pre-training to RGB-D network vs training from scratch (ablation) | U-Net RGB-D · SUN RGB-D (13 class, val) | mean IU: 41.4 -> 49.8 | Improved |
Closest papers by meaning
- ScanNet: Richly-Annotated 3D Reconstructions of Indoor Scenes (CVPR 2017 · similarity 0.91)
- Physically-Based Rendering for Indoor Scene Understanding Using Convolutional Neural Networks (CVPR 2017 · similarity 0.91)
- Understanding Real World Indoor Scenes With Synthetic Data (CVPR 2016 · similarity 0.90)
- ScanNet++: A High-Fidelity Dataset of 3D Indoor Scenes (ICCV 2023 · similarity 0.89)
- DFormer: Rethinking RGBD Representation Learning for Semantic Segmentation (ICLR 2024 · similarity 0.88)
- ImageNet-D: Benchmarking Neural Network Robustness on Diffusion Synthetic Object (CVPR 2024 · 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.
