Research map / 3D vision / Depth and stereo
RDFC-GAN: RGB-Depth Fusion CycleGAN for Indoor Depth Completion
Haowen Wang, Zhengping Che, Yufan Yang, Mingyuan Wang, Zhiyuan Xu, Xiuquan Qiao, Mengshi Qi, Feifei Feng, Jian Tang
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 CycleGAN consistency over a plain GAN (ablation) | RDFC-GAN · NYU-Depth V2 (Setting A) | RMSE: 0.129 -> 0.120 | Improved |
| add a GAN branch to the dual encoder-decoder (ablation) | RDFC-GAN · NYU-Depth V2 (Setting A) | RMSE: 0.176 -> 0.129 | Improved |
| fine-tune the pretrained normal generator on NYU-Depth V2 (ablation) | MCN branch · NYU-Depth V2 (Setting A) | RMSE: 0.147 -> 0.132 | Improved |
| add Manhattan constraint losses via segmentation network (MNM) over fine-tuned normal generator (ablation) | MCN branch · NYU-Depth V2 (Setting A) | RMSE: 0.132 -> 0.120 | Improved |
| replace Local Guidance with Manhattan Normal Module (ablation) | RDFC-GAN · NYU-Depth V2 (Setting A) | RMSE: 0.146 -> 0.120 | Improved |
| add segmentation features as extra input to the normal generator (ablation) | Manhattan Normal Module · NYU-Depth V2 (Setting A) | RMSE: 0.120 -> 0.122 | Hurt |
| add wall loss to normal loss (ablation) | MCN branch · NYU-Depth V2 (Setting A) | RMSE: 0.132 -> 0.123 | Improved |
| add floor, ceiling and wall losses together (ablation) | MCN branch · NYU-Depth V2 (Setting A) | RMSE: 0.132 -> 0.120 | Improved |
| use W-AdaIN fusion instead of instance normalization (ablation) | RDFC-GAN · NYU-Depth V2 (Setting A) | RMSE: 0.126 -> 0.120 | Improved |
| use W-AdaIN fusion instead of plain AdaIN (ablation) | RDFC-GAN · NYU-Depth V2 (Setting A) | RMSE: 0.133 -> 0.120 | Improved |
| feed RDFC-GAN completed depth to detector instead of original input | VoteNet · SUN RGB-D | mAP@25: 59.07 -> 61.02 | Improved |
| feed NLSPN completed depth to detector | H3DNet · SUN RGB-D | mAP@25: 60.11 -> 27.10 | Hurt |
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.
- GraphCSPN — GraphCSPN: Geometry-Aware Depth Completion via Dynamic GCNs (ECCV 2022) · better on 4 of 5 shared results
- DeepLidar — DeepLiDAR: Deep Surface Normal Guided Depth Prediction for Outdoor Scene From Sparse LiDAR Data and Single Color Image (CVPR 2019) · better on 1 of 1 shared result
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
- RGB-Depth Fusion GAN for Indoor Depth Completion (CVPR 2022 · similarity 0.97)
- Estimating Depth from RGB and Sparse Sensing (ECCV 2018 · similarity 0.89)
- Deep Depth Completion of a Single RGB-D Image (CVPR 2018 · similarity 0.88)
- Efficient Indoor Depth Completion Network Using Mask-adaptive Gated Convolution (AAAI 2025 · similarity 0.87)
- PatchFusion: An End-to-End Tile-Based Framework for High-Resolution Monocular Metric Depth Estimation (CVPR 2024 · similarity 0.86)
- Bilateral Propagation Network for Depth Completion (CVPR 2024 · similarity 0.86)
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.
