Research map / Image generation / Super-resolution
Face Hallucination by Attentive Sequence Optimization with Reinforcement Learning
Yukai Shi, Guanbin Li, Qingxing Cao, Keze Wang, Liang Lin
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 |
|---|---|---|---|
| attend patches with the RL policy instead of enhancing the whole face (ablation) | Attention-FH · LFW (x4) | PSNR: +0.73 dB | Improved |
| attend patches with the RL policy instead of enhancing the whole face (ablation) | Attention-FH · LFW (x8) | PSNR: +0.28 dB | Improved |
| use RL policy instead of random patch selection (ablation) | Attention-FH · LFW (x4) | PSNR: 31.63 -> 33.02 | Improved |
| use RL policy instead of sequential scanning (ablation) | Attention-FH · LFW (x8) | PSNR: 25.98 -> 26.24 | Improved |
| use RL policy instead of spatial transformer network (ablation) | Attention-FH · LFW (x4) | PSNR: 28.13 -> 33.02 | Improved |
| feed restored image I_t instead of I_0 to the policy network (ablation) | Attention-FH · LFW (x4) | PSNR: 32.12 -> 33.02 | Improved |
| use size-free attention instead of fixed 60x60 box (ablation) | Attention-FH · LFW (x4) | PSNR: +0.13 dB | Improved |
| use size-free attention instead of fixed 60x60 box (ablation) | Attention-FH · LFW (x8) | PSNR: +0.12 dB | Improved |
| use random-patch recurrent enhancement instead of plain CNN-16 (ablation) | Attention-FH · LFW (x4) | PSNR: 29.11 -> 31.63 | Improved |
| increase recurrent steps T beyond 15 (ablation) | Attention-FH · LFW | PSNR | No clear effect |
| use PSNR-gain stabilized reward instead of absolute PSNR reward (ablation) | Attention-FH · LFW | PSNR | Improved |
| use EDSR as local enhancement network (ablation) | Attention-FH · LFW (x8) | PSNR: 27.81 -> 28.25 | Improved |
Closest papers by meaning
- Dynamic Multimodal Activation Steering for Hallucination Mitigation in Large Vision-Language Models (ICLR 2026 · similarity 0.86)
- Locate-then-Sparsify: Attribution Guided Sparse Strategy for Visual Hallucination Mitigation (CVPR 2026 · similarity 0.86)
- Steer Where It Matters: Token-Level Visual-Sensitivity Steering for LVLMs Hallucination Mitigation (ICML 2026 · similarity 0.86)
- P$^2$-DPO:Grounding Hallucination in Perceptual Processing via Calibration Direct Preference Optimization (ICLR 2026 · similarity 0.86)
- SAVER: Mitigating Hallucinations in Large Vision-Language Models via Style-Aware Visual Early Revision (AAAI 2026 · similarity 0.86)
- Hallucination-aware Intermediate Representation Editing in Large Vision-Lanugage Models (ICLR 2026 · similarity 0.86)
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.
