Research map / Image generation / GANs
OneGAN: Simultaneous Unsupervised Learning of Conditional Image Generation, Foreground Segmentation, and Fine-Grained Clustering
Yaniv Benny, Lior Wolf
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 multi-phase training schedule (ablation) | OneGAN · Birds segmentation | IOU: 28.2 -> 55.5 | Improved |
| add multi-phase training schedule (ablation) | OneGAN · Birds generation | FID: 196.5 -> 20.5 | Improved |
| add reconstruction path (Phase II) to generation-only training (ablation) | OneGAN · Birds segmentation | IOU: 45.7 -> 55.5 | Improved |
| add reconstruction of real images (ablation) | OneGAN · Birds segmentation | IOU: 53.5 -> 55.5 | Improved |
| add reconstruction of real images (ablation) | OneGAN · Birds generation | CIS: 25.6 -> 30.7 | Improved |
| add reconstruction of real images (ablation) | OneGAN · Cars clustering | ACC: .057 -> .060 | Mixed |
| use GLU layer normalization instead of layer normalization (ablation) | OneGAN generators · CUB | FID: 87.5 -> 20.5 | Improved |
| use GLU layer normalization instead of instance normalization (ablation) | OneGAN generators · CUB | IOU: 30.1 -> 55.5 | Improved |
| add mixup at embedding level (mixup 1,2) (ablation) | OneGAN · CUB | C-IS: 17.5 -> 30.7 | Improved |
| add mixup at skip connections (mixup 3,4) (ablation) | OneGAN · CUB | C-IS: 22.2 -> 30.7 | Improved |
| add bypass skip connection (ablation) | OneGAN · CUB | FID: 21.2 -> 20.5 | No clear effect |
| add mask regularization loss L_RM (ablation) | OneGAN · CUB | FID: 97.2 -> 20.5 | Improved |
Later papers that compared with this method
1 later accepted paper reports OneGAN as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Foreground-Background Separation through Concept Distillation from Generative Image Foundation Models (ICCV 2023 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- StackGANv2 — StackGAN++: Realistic Image Synthesis with Stacked Generative Adversarial Networks (TPAMI 2018) · better on 2 of 2 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
- Finding an Unsupervised Image Segmenter in each of your Deep Generative Models (ICLR 2022 · similarity 0.89)
- Object Segmentation Without Labels with Large-Scale Generative Models (ICML 2021 · similarity 0.88)
- Foreground-Background Separation through Concept Distillation from Generative Image Foundation Models (ICCV 2023 · similarity 0.88)
- ComGAN: Unsupervised Disentanglement and Segmentation via Image Composition (NeurIPS 2022 · similarity 0.88)
- A Generative Approach at the Instance-Level for Image Segmentation Under Limited Training Data Conditions (Student Abstract) (AAAI 2025 · similarity 0.85)
- Segment Anything without Supervision (NeurIPS 2024 · similarity 0.85)
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.
