Research map / Vision-language and multimodal / Human and embodied modeling
Aesthetically Relevant Image Captioning
Zhipeng Zhong, Fei Zhou, Guoping Qiu
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 |
|---|---|---|---|
| Weight caption training loss by sentence ARS rather than using standard cross-entropy (ablation) | VisualGPT image aesthetic captioning on DPC2022 · DPC2022 image aesthetic captioning | SPICE: 0.0172 -> 0.0353 | Improved |
| Weight caption training loss by sentence ARS rather than using standard cross-entropy (ablation) | VisualGPT image aesthetic captioning on DPC2022 · DPC2022 image aesthetic captioning | METETOR: 0.1227 -> 0.1389 | Improved |
| Select generated captions with ARS-based DACS instead of using captions without aesthetic selection (ablation) | Generated-caption text-only AQA · DPC2022 generated-caption text-only AQA | SRCC: 0.4335 -> 0.4711 | Improved |
| Select generated captions with CLIP-based DACS instead of using captions without aesthetic selection (ablation) | Generated-caption text-only AQA · DPC2022 generated-caption text-only AQA | SRCC: 0.4335 -> 0.476 | Improved |
| Use ARS-selected captions instead of unselected generated captions (ablation) | Generated-caption multi-modal AQA · DPC2022 generated-caption multi-modal AQA | MAE: 0.4232 -> 0.4192 | Improved |
| Use CLIP-selected captions instead of unselected generated captions (ablation) | Generated-caption multi-modal AQA · DPC2022 generated-caption multi-modal AQA | MAE: 0.4232 -> 0.4179 | Improved |
| Use captions selected by DACS instead of generator-confidence captions (ablation) | Generated-caption multi-modal AQA · DPC2022 generated-caption multi-modal AQA | ACC: 0.8181 -> 0.8203 | Improved |
| Use higher-ARS comments rather than lower-ARS comments (ablation) | ViT+BERT multi-modal AQA · DPC2022 multi-modal AQA | SRCC: 0.8216 -> 0.9157 | Improved |
| Use higher-ARS comments rather than lower-ARS comments (ablation) | ViT+BERT multi-modal AQA · DPC2022 multi-modal AQA | ACC: 0.8486 -> 0.8991 | Improved |
| Use generated captions rather than ground-truth text (ablation) | Text-only AQA · DPC2022 text-only AQA | SRCC: 0.8274 -> 0.476 | Hurt |
Closest papers by meaning
- Aesthetic Critiques Generation for Photos (ICCV 2017 · similarity 0.89)
- CaptionQA: Is Your Caption as Useful as the Image Itself? (CVPR 2026 · similarity 0.86)
- Advancing Comprehensive Aesthetic Insight with Multi-Scale Text-Guided Self-Supervised Learning (AAAI 2025 · similarity 0.85)
- ArtiMuse: Fine-Grained Image Aesthetics Assessment with Joint Scoring and Expert-Level Understanding (CVPR 2026 · similarity 0.85)
- Bridging Cognitive Gap: Hierarchical Description Learning for Artistic Image Aesthetics Assessment (AAAI 2026 · similarity 0.83)
- Aligning Vision Models with Human Aesthetics in Retrieval: Benchmarks and Algorithms (NeurIPS 2024 · similarity 0.83)
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.
