Research map / Image generation / Editing and style transfer
AugGAN: Cross Domain Adaptation with GAN-based Data Augmentation
Sheng-Wei Huang, Che-Tsung Lin, Shu-Ping Chen, Yen-Yi Wu, Po-Hao Hsu, Shang-Hong Lai
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 segmentation parsing to the inverse cycle as well as the forward cycle (ablation) | AugGAN · SYNTHIA day-train, SYNTHIA night-test (Faster RCNN) | vehicle detection AP (%): 68.7 -> 72.2 | Improved |
| add segmentation parsing to the inverse cycle as well as the forward cycle (ablation) | AugGAN · SYNTHIA day-train, SYNTHIA night-test (YOLO) | vehicle detection AP (%): 38.1 -> 39.0 | Improved |
| add soft weight sharing (lambda_w=0.02) in the decoder instead of hard sharing (ablation) | AugGAN · SYNTHIA day-train, SYNTHIA night-test (Faster RCNN) | vehicle detection AP (%): 57.2 -> 68.7 | Improved |
| share decoder weights between generator and parsing net (hard) on top of encoder hard-sharing (ablation) | AugGAN · SYNTHIA day-train, SYNTHIA night-test (Faster RCNN) | vehicle detection AP (%): 39.9 -> 57.2 | Improved |
| use AugGAN instead of CycleGAN for translated training data | Faster RCNN · SYNTHIA day-train, SYNTHIA night-test | vehicle detection AP (%): 65.9 -> 72.2 | Improved |
| use AugGAN instead of UNIT for translated training data | YOLO · KITTI-D2N (GTA GAN) train, ITRI-Night test | vehicle detection AP (%): 20.5 -> 46.0 | Improved |
| use AugGAN instead of CycleGAN on GTA-trained translation | Faster-RCNN · GTA day-to-night train, GTA night-test | vehicle detection AP (%): 54.4 -> 67.4 | Improved |
| use AugGAN instead of CycleGAN on ITRI-Day with GTA-trained GAN | Faster RCNN · ITRI-Day D2N (GTA GAN) train, ITRI-Night test | vehicle detection AP (%): 86.2 -> 86.1 | No clear effect |
| use AugGAN instead of CycleGAN on ITRI-Day with SYNTHIA-trained GAN | Faster RCNN · ITRI-Day D2N (SYNTHIA GAN) train, ITRI-Night test | vehicle detection AP (%): 72.4 -> 81.2 | Improved |
| use AugGAN instead of CycleGAN on ITRI-Day with GTA-trained GAN | YOLO · ITRI-Day D2N (GTA GAN) train, ITRI-Night test | vehicle detection AP (%): 37.9 -> 44.1 | Improved |
Later papers that compared with this method
1 later accepted paper reports AugGAN as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Multimodal Structure-Consistent Image-to-Image Translation (AAAI 2020 · better on 6 of 6 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
- TraVeLGAN: Image-To-Image Translation by Transformation Vector Learning (CVPR 2019 · similarity 0.92)
- ReMix: Towards Image-to-Image Translation With Limited Data (CVPR 2021 · similarity 0.91)
- Unsupervised Image-to-Image Translation Networks (NeurIPS 2017 · similarity 0.91)
- DeepI2I: Enabling Deep Hierarchical Image-to-Image Translation by Transferring from GANs (NeurIPS 2020 · similarity 0.90)
- E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 2024 · similarity 0.88)
- Rethinking the Paradigm of Content Constraints in Unpaired Image-to-Image Translation (AAAI 2024 · similarity 0.88)
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.
