Research map / Image generation / Autoregressive and latent generation
Finding Highlight Images In Your Albums:From Benchmark To MLLM
Rong Qin, Congcong Sun, Yaopeng Dong, Yanbin Sun, Chensen Ding, Chenxi Zhao, Biao Wang, Qian Zhang, Eunil Park, Chi Man VONG, Jufeng Yang
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 supervised fine-tuning on AHR labels (ablation) | Qwen3-VL-8B · PEC (Top-5%) | Precision@5%: 28.6 -> 66.9 | Improved |
| add training-free HDA on top of fine-tuning (ablation) | Fine-tuned Qwen3-VL-8B · PEC (Top-5%) | Precision@5%: 66.9 -> 73.4 | Improved |
| add MLP-based HDA* on top of fine-tuning (ablation) | Fine-tuned Qwen3-VL-8B · PEC (Top-5%) | Precision@5%: 66.9 -> 73.5 | Improved |
| use MLP HDA* instead of training-free HDA in full model (ablation) | Highlight4U · PEC (Top-5%) | Precision@5%: 74.1 -> 74.4 | No clear effect |
| add image-only global references (ablation) | Fine-tune + HDA · PEC (Top-15%) | Precision@15%: 58.4 -> 60.0 | Improved |
| add reference captions (Ref-Text) to image references (ablation) | Fine-tune + HDA + Ref-Image · PEC (Top-15%) | Precision@15%: 60.0 -> 61.4 | Improved |
| add global reference re-recommendation (ablation) | Fine-tune + HDA · PEC (Top-15%) | Recall@15%: 56.6 -> 60.0 | Improved |
| add image-only references at top 5% (ablation) | Fine-tune + HDA · PEC (Top-5%) | Precision@5%: 73.4 -> 73.3 | No clear effect |
| reference count M too small (1 or 2) (ablation) | Highlight4U · PEC | Precision/Recall | Mixed |
| reference count M too large (9) (ablation) | Highlight4U · PEC | Precision/Recall | Hurt |
| use regression head instead of binary classification head | ConvNeXt-B · PEC (Top-5%) | Precision@5%: 66.9 -> 68.8 | Improved |
| use CLIP instead of DINOv2 as frozen backbone | ViT-B linear probe · PEC (Top-5%) | Precision@5%: 58.5 -> 71.6 | Improved |
Closest papers by meaning
- Beyond Single Images: A Comprehensive Benchmark for Album-Level Vision-Language Understanding (CVPR 2026 · similarity 0.86)
- Event-Specific Image Importance (CVPR 2016 · similarity 0.83)
- RAGAR: Retrieval Augmented Personalized Image Generation Guided by Recommendation (AAAI 2026 · similarity 0.83)
- PR-Net: Preference Reasoning for Personalized Video Highlight Detection (ICCV 2021 · similarity 0.82)
- Tell Me What Is Good about This Property: Leveraging Reviews for Segment-Personalized Image Collection Summarization (AAAI 2024 · similarity 0.82)
- Stylebreeder: Exploring and Democratizing Artistic Styles through Text-to-Image Models (NeurIPS 2024 · similarity 0.81)
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.
