Research map / Image generation / GANs
Deep Generative Adversarial Compression Artifact Removal
Leonardo Galteri, Lorenzo Seidenari, Marco Bertini, Alberto Del Bimbo
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 the GAN with perceptual loss to the residual generator for object detection | Residual generator · PASCAL VOC2007 detection (Faster R-CNN), JPEG QF=20 restored | mAP: 0.574 -> 0.623 | Improved |
| train with SSIM loss instead of MSE | Residual generator · LIVE1, JPEG QF=10 | SSIM: 0.834 -> 0.840 | Improved |
| train with SSIM loss instead of MSE | Residual generator · LIVE1, JPEG QF=10 | PSNR (dB): 29.45 -> 28.94 | Hurt |
| add adversarial training (GAN) instead of MSE | Residual generator · LIVE1, JPEG QF=10 | PSNR (dB): 29.45 -> 27.29 | Hurt |
| use GAN instead of SSIM loss for perceived quality | Residual generator · BSD500 subjective study | MOS: 49.51 -> 68.32 | Improved |
| use perceptual loss l_P instead of simple L1 loss in GAN (ablation) | Our GAN · PASCAL VOC2007 detection (Faster R-CNN), JPEG QF=20 restored | mAP: 0.598 -> 0.623 | Improved |
| add color (RGB) instead of luminance only in GAN (ablation) | Our GAN · PASCAL VOC2007 detection (Faster R-CNN), JPEG QF=20 restored | mAP: 0.598 -> 0.623 | Improved |
| use sub-patch discriminator instead of full patch discriminator (ablation) | Our GAN · PASCAL VOC2007 detection (Faster R-CNN), JPEG QF=20 restored | mAP: 0.605 -> 0.623 | Improved |
| restore compressed images with MSE/SSIM generator before detection | Faster R-CNN · PASCAL VOC2007, classes cat, cow, dog, sheep | per-class AP | Hurt |
| apply GAN restoration to WebP-compressed images | Faster R-CNN · PASCAL VOC2007, WebP | mAP: +4% | Improved |
| apply GAN restoration to JPEG2000 images | Faster R-CNN · PASCAL VOC2007, JPEG2000 | mAP: +2.9% | Improved |
| apply GAN restoration to BPG images | Faster R-CNN · PASCAL VOC2007, BPG | mAP: +2.1% | Improved |
Closest papers by meaning
- Generative Adversarial Networks for Extreme Learned Image Compression (ICCV 2019 · similarity 0.88)
- How To Exploit the Transferability of Learned Image Compression to Conventional Codecs (CVPR 2021 · similarity 0.87)
- High-Fidelity Generative Image Compression (NeurIPS 2020 · similarity 0.87)
- On Aliased Resizing and Surprising Subtleties in GAN Evaluation (CVPR 2022 · similarity 0.87)
- Physics-Based Generative Adversarial Models for Image Restoration and Beyond (TPAMI 2020 · similarity 0.87)
- Unsupervised Class-Specific Deblurring (ECCV 2018 · 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.
