Research map / Image generation / Super-resolution
Super-Identity Convolutional Neural Network for Face Hallucination
Kaipeng Zhang, Zhanpeng Zhang, Chia-Wen Cheng, Winston H. Hsu, Yu Qiao, Wei Liu, Tong Zhang
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 super-identity loss (alpha 0 -> 8) (ablation) | SICNN · Set C (12x14 -> 96x112, 8x) | identity similarity: 0.4418 -> 0.5978 | Improved |
| add super-identity loss (alpha 0 -> 8) (ablation) | SICNN · LFW | accuracy: 97.61% -> 98.25% | Improved |
| add super-identity loss (alpha 0 -> 8) (ablation) | SICNN · YTF | accuracy: 93.20% -> 93.82% | Improved |
| increase alpha from 8 to 32 (ablation) | SICNN · LFW | accuracy: 98.25% -> 98.16% | No clear effect |
| use domain-integrated training instead of joint training from scratch (approach I) (ablation) | SICNN · Set C (12x14 -> 96x112, 8x) | identity similarity: 0.3875 -> 0.5978 | Improved |
| use domain-integrated training instead of fixed HR-trained recognizer (approach II) (ablation) | SICNN · Set C (12x14 -> 96x112, 8x) | identity similarity: 0.4829 -> 0.5978 | Improved |
| use domain-integrated training instead of one-shot alternating training (approach III) (ablation) | SICNN · Set C (12x14 -> 96x112, 8x) | identity similarity: 0.5132 -> 0.5978 | Improved |
| use domain-integrated training instead of approach III (ablation) | SICNN · LFW | accuracy: 97.58% -> 98.25% | Improved |
| add perceptual loss from recognition net to UR-DGN (UR-DGN*) | UR-DGN · Set C (12x14 -> 96x112, 8x) | identity similarity: 0.3682 -> 0.5267 | Improved |
| SICNN versus UR-DGN* (adding super-identity loss with DI training) | UR-DGN* · Set C (12x14 -> 96x112, 8x) | identity similarity: 0.5267 -> 0.5978 | Improved |
| use adversarial loss (UR-DGN) instead of L2 only (LapSRN) | LapSRN · Set C (12x14 -> 96x112, 8x) | identity similarity: 0.4361 -> 0.3682 | Hurt |
| use SICNN instead of LapSRN | LapSRN · Set C (12x14 -> 96x112, 8x) | PSNR (dB): 26.1451 -> 26.8945 | Improved |
Closest papers by meaning
- Semantic Face Hallucination: Super-Resolving Very Low-Resolution Face Images with Supplementary Attributes (TPAMI 2019 · similarity 0.91)
- Face Hallucination with Tiny Unaligned Images by Transformative Discriminative Neural Networks (AAAI 2017 · similarity 0.90)
- Super-Resolving Very Low-Resolution Face Images With Supplementary Attributes (CVPR 2018 · similarity 0.89)
- Can We See More? Joint Frontalization and Hallucination of Unaligned Tiny Faces (TPAMI 2019 · similarity 0.89)
- Deep Cascaded Bi-Network for Face Hallucination (ECCV 2016 · similarity 0.89)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 2025 · similarity 0.84)
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.
