Research map / Vision-language and multimodal / CLIP and zero-shot
VLIC: Vision-Language Models As Perceptual Judges for Human-Aligned Image Compression
Kyle Sargent, Ruiqi Gao, Philipp Henzler, Charles Herrmann, Aleksander Holynski, Li Fei-Fei, Jiajun Wu, Jason Y. 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 Diffusion DPO post-training with VLM+LPIPS reward (ablation) | VLIC · MS-COCO | FD-DINO: 82.31 -> 67.83 | Improved |
| add VLM reward to LPIPS-only post-training (0.07 bpp) (ablation) | VLIC · MS-COCO | Human Elo: 838 -> 858 | Improved |
| add VLM reward to LPIPS-only post-training (0.21 bpp) (ablation) | VLIC · MS-COCO | Human Elo: 1103 -> 1112 | Improved |
| add VLM reward to LPIPS-only post-training (0.21 bpp) (ablation) | VLIC · MS-COCO | FID: 1.30 -> 1.35 | No clear effect |
| ensemble VLM reward with LPIPS (ablation) | VLIC · MS-COCO | FD-DINO: 67.68 -> 67.83 | No clear effect |
| ensemble VLM reward with LPIPS (ablation) | VLIC · MS-COCO | PSNR: 21.29 -> 21.68 | Improved |
| add VLM self-ensembling over 3 seeds (ablation) | VLIC · MS-COCO | FD-DINO: 68.36 -> 67.83 | Improved |
| add VLM self-ensembling over 3 seeds (ablation) | VLIC · MS-COCO | FID: 2.15 -> 2.31 | Mixed |
| increase number of VLM random seeds (ablation) | VLM judge · BAPPS | 2AFC accuracy: 67% to about 71% (Figure 5) | Improved |
| use Gemini 2.5-Flash as zero-shot judge instead of LPIPS | VLM judge · BAPPS-Val | 2AFC accuracy: 69.56 vs 69.44 | No clear effect |
| use VLM judge instead of LPIPS on compressed images | VLM judge · Compressed images | 2AFC accuracy: 92.32 vs 83.80 | Hurt |
Closest papers by meaning
- Benchmarking and Enhancing VLM for Compressed Image Understanding (ICML 2026 · similarity 0.90)
- Towards Unified Human Perception and Machine Understanding: Token Flow Guided Compression Framework (CVPR 2026 · similarity 0.90)
- LLMC+: Benchmarking Vision-Language Model Compression with a plug-and-play Toolkit (AAAI 2026 · similarity 0.89)
- DUET-VLM: Dual stage Unified Efficient Token reduction for VLM Training and Inference (CVPR 2026 · similarity 0.88)
- Unified Coding for Both Human Perception and Generalized Machine Analytics with CLIP Supervision (AAAI 2025 · similarity 0.88)
- Bridging Compressed Image Latents and Multimodal Large Language Models (ICLR 2025 · 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.
