Research map / Image generation / Editing and style transfer
ForkGAN: Seeing into the Rainy Night
Ziqiang Zheng, Yang Wu, Xinran Han, Jianbo Shi
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 fork-shape generator (one encoder, two decoders) (ablation) | ForkGAN · Alderley | FID / SIFT: 146 / 4.26 -> 131 / 7.12 | Improved |
| add dilated residual blocks (ablation) | Fork-shape generator · Alderley | FID / SIFT: 131 / 7.12 -> 113 / 8.12 | Improved |
| add multi-scale discriminator (ablation) | Fork-shape+DRB · Alderley | FID / SIFT: 113 / 8.12 -> 95.3 / 9.14 | Improved |
| add perceptual loss Lper and domain classification loss Lcls (ablation) | Fork-shape+DRB+MSD · Alderley | FID / SIFT: 95.3 / 9.14 -> 70.7 / 11.5 | Improved |
| add refinement stage at test time (ablation) | ForkGAN · Alderley | FID / SIFT: 70.7 / 11.5 -> 61.2 / 12.1 | Improved |
| use separate translation and reconstruction decoders instead of shared decoders (ablation) | ForkGAN · Alderley | FID / SIFT: 73.8 / 9.29 -> 61.2 / 12.1 | Improved |
| night-to-day translation with ForkGAN before segmentation | DeepLabv3 · BDD100K night (137 images) | mIoU: 7.03 -> 14.4 | Improved |
| night-to-day translation with ToDayGAN before segmentation | DeepLabv3 · BDD100K night (137 images) | mIoU: 7.03 -> 8.19 | Improved |
| night-to-day translation with MUNIT before segmentation | DeepLabv3 · BDD100K night (137 images) | mIoU: 7.03 -> 2.44 | Hurt |
| augment day-only training with ForkGAN night translations | Faster R-CNN · BDD100K night detection | mAP: 22.1 -> 22.9 | Improved |
| augment day-only training with ToDayGAN translations | Faster R-CNN · BDD100K night detection | mAP: 22.1 -> 19.5 | Hurt |
| add ForkGAN translated night images to real night training data | Faster R-CNN · BDD100K night detection | mAP: 23.9 -> 26.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- DRIT — DRIT++: Diverse Image-to-Image Translation via Disentangled Representations (IJCV 2020) · better on 2 of 2 shared results
- UGATIT — U-GAT-IT: Unsupervised Generative Attentional Networks with Adaptive Layer-Instance Normalization for Image-to-Image Translation (ICLR 2020) · better on 1 of 1 shared result
- StarGAN — StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation (CVPR 2018) · 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
- Weatherproofing Retrieval for Localization with Generative AI and Geometric Consistency (ICLR 2024 · similarity 0.86)
- Parsing All Adverse Scenes: Severity-Aware Semantic Segmentation with Mask-Enhanced Cross-Domain Consistency (AAAI 2024 · similarity 0.84)
- Harnessing Joint Rain-/Detail-aware Representations to Eliminate Intricate Rains (ICLR 2024 · similarity 0.84)
- Semantic Segmentation on Raindrop Degraded Images Using Two-Stage Dual Teacher-Student Learning (AAAI 2025 · similarity 0.84)
- End-to-End Video Semantic Segmentation in Adverse Weather using Fusion Blocks and Temporal-Spatial Teacher-Student Learning (NeurIPS 2024 · similarity 0.83)
- WeatherPrompt: Multi-modality Representation Learning for All-Weather Drone Visual Geo-Localization (NeurIPS 2025 · 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.
