Research map / Detection and segmentation / Open-set and few-shot detection
SeGAN: Segmenting and Generating the Invisible
Kiana Ehsani, Roozbeh Mottaghi, Ali Farhadi
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 |
|---|---|---|---|
| Jointly predict the full-object mask and paint its invisible regions | SeGAN · DYCE synthetic test, full-object segmentation | IoU: 47.51 -> 68.78 | Improved |
| Jointly predict the full-object mask and paint its invisible regions | SeGAN · DYCE synthetic test, invisible-region segmentation | IoU: 12.3 -> 15.59 | Improved |
| Use SeGAN instead of Pix2Pix for appearance generation | DYCE synthetic test · DYCE synthetic test | L1 distance: 0.15 -> 0.07 | Improved |
| Use SeGAN instead of Pix2Pix for appearance generation | DYCE synthetic test · DYCE synthetic test | L2 distance: 0.09 -> 0.03 | Improved |
| Train on synthetic occlusion data and evaluate on natural images | SeGAN · PASCAL 3D indoor categories, invisible-region segmentation | IoU: 13.9 -> 18.7 | Improved |
| Combine natural and synthetic training data instead of using natural data alone | Multipath · PASCAL 3D indoor categories, invisible-region segmentation | IoU: 13.9 -> 8.4 | Hurt |
| Use groundtruth visible-region masks instead of predicted masks | SeGAN · DYCE synthetic test, invisible-region segmentation | IoU: 15.59 -> 23.26 | Improved |
| Use groundtruth visible-region masks instead of predicted masks | SeGAN · DYCE synthetic test | L1 distance: 0.07 -> 0.03 | Improved |
| Infer object occlusion from predicted invisible-region masks rather than single-image depth estimates | SeGAN · DYCE test occlusion pairs | depth layering accuracy: 60.18 -> 84.04 | Improved |
Later papers that compared with this method
2 later accepted papers report SeGAN as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Visiting the Invisible: Layer-by-Layer Completed Scene Decomposition (IJCV 2021 · better on 1 of 1 shared result)
- Visualizing the Invisible: Occluded Vehicle Segmentation and Recovery (ICCV 2019 · better on 4 of 4 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
- Visiting the Invisible: Layer-by-Layer Completed Scene Decomposition (IJCV 2021 · similarity 0.88)
- RevealNet: Seeing Behind Objects in RGB-D Scans (CVPR 2020 · similarity 0.85)
- Visualizing the Invisible: Occluded Vehicle Segmentation and Recovery (ICCV 2019 · similarity 0.84)
- Using Diffusion Priors for Video Amodal Segmentation (CVPR 2025 · similarity 0.84)
- Completing Visual Objects via Bridging Generation and Segmentation (ICML 2024 · similarity 0.83)
- Amodal Scene Analysis via Holistic Occlusion Relation Inference and Generative Mask Completion (AAAI 2024 · similarity 0.83)
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.
