Research map / Image generation / Editing and style transfer
Model-based occlusion disentanglement for image-to-image translation
Fabio Pizzati, Pietro Cerri, Raoul de Charette
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 adversarial disentanglement with injected physical raindrop model to MUNIT backbone | MUNIT · nuScenes clear-to-rain | IS: 1.211 -> 1.532 | Improved |
| add adversarial disentanglement with injected physical raindrop model to MUNIT backbone | MUNIT · nuScenes clear-to-rain | LPIPS diversity: 0.495 -> 0.515 | Improved |
| add adversarial disentanglement with injected physical raindrop model to MUNIT backbone | MUNIT · nuScenes clear-to-rain | CIS: 1.030 -> 1.148 | Improved |
| use our translated rainy Cityscapes images instead of physics-based rain rendering for finetuning | PSPNet · Rainy nuScenes semantic segmentation | AP: 25.6 -> 27.7 | Improved |
| finetune with our rainy images vs original weights | PSPNet · Rainy nuScenes semantic segmentation | AP: 18.7 -> 27.7 | Improved |
| use the full raindrop occlusion model instead of the simpler Refract or Gaussian models (ablation) | Ours · nuScenes clear-to-rain | GAN metrics | Improved |
| use the matching occlusion model (drop) rather than watermark, dirt or fence models (ablation) | Ours · RobotCar real raindrops | FID: 329.17 -> 135.32 | Improved |
| use the drop model instead of the fence model (ablation) | Ours · RobotCar real raindrops | FID: 948.71 -> 135.32 | Improved |
| add disentanglement guidance with beta 0.75 versus no guidance (beta 1) (ablation) | Ours · nuScenes clear-to-rain | qualitative | Improved |
| set guidance beta to 0 (no occlusions injected) versus beta 0.75 (ablation) | Ours · nuScenes clear-to-rain | qualitative | Hurt |
| use estimated defocus sigma 3.87 versus other tested sigma values (ablation) | Ours · RobotCar | FID | Improved |
| use our parameter estimation for rendered drops instead of prior drop rendering | Ours · RobotCar | FID and LPIPS distance | Improved |
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.
- DRIT — DRIT++: Diverse Image-to-Image Translation via Disentangled Representations (IJCV 2020) · better on 3 of 3 shared results
- U-GAT-IT — U-GAT-IT: Unsupervised Generative Attentional Networks with Adaptive Layer-Instance Normalization for Image-to-Image Translation (ICLR 2020) · 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
- Scaling-Up Disentanglement for Image Translation (ICCV 2021 · similarity 0.89)
- Physics-Informed Guided Disentanglement in Generative Networks (TPAMI 2023 · similarity 0.88)
- Multi-mapping Image-to-Image Translation via Learning Disentanglement (NeurIPS 2019 · similarity 0.88)
- TransGaGa: Geometry-Aware Unsupervised Image-To-Image Translation (CVPR 2019 · similarity 0.88)
- Image-to-image translation for cross-domain disentanglement (NeurIPS 2018 · similarity 0.87)
- OOGAN: Disentangling GAN with One-Hot Sampling and Orthogonal Regularization (AAAI 2020 · 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.
