Research map / Vision-language and multimodal / CLIP and zero-shot
Positive-Augmented Contrastive Learning for Vision-and-Language Evaluation and Training
Sara Sarto, Nicholas Moratelli, Marcella Cornia, Lorenzo Baraldi, Rita Cucchiara
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 LoRA fine-tuning instead of tuning only last projections (ablation) | PAC-S++ (ViT-B/32) · Composite | Kendall tau_c: 58.3 on the Composite dataset (+2.6) | Improved |
| add LoRA fine-tuning instead of tuning only last projections (ablation) | PAC-S++ (ViT-B/32) · VATEX-EVAL | Kendall tau_b: 28.1 on the VATEX-EVAL dataset (+3.0) | Improved |
| add generated visual and textual positives (ablation) | PAC-S++ (ViT-B/32) · Flickr8k-Expert | Kendall tau_c: 53.9 -> 54.5 | Improved |
| add generated visual and textual positives (ablation) | PAC-S++ (ViT-B/32) · Composite | Kendall tau_c: 57.5 -> 58.3 | Improved |
| add generated visual and textual positives (ablation) | PAC-S++ (ViT-B/32) · VATEX-EVAL | Kendall tau_b: 26.7 -> 28.1 | Improved |
| add synthetic captions (BLIP) to synthetic images (ablation) | PAC-S++ (ViT-B/32) · Composite | Kendall tau_c: 57.9 -> 58.3 | Improved |
| add synthetic images (SDv1.5) to synthetic captions (ablation) | PAC-S++ (ViT-B/32) · Composite | Kendall tau_c: 57.3 -> 58.3 | Improved |
| raise lambda_v from 0.1 to 0.5 (ablation) | PAC-S++ (ViT-B/32) · Composite | Kendall tau_c: 58.3 -> 57.9 | Hurt |
| raise lambda_t from 0.001 to 0.1 (ablation) | PAC-S++ (ViT-B/32) · Composite | Kendall tau_c: 58.3 -> 57.5 | Hurt |
| replace BLIP with LLaMA-3.2 as caption generator (ablation) | PAC-S++ (ViT-B/32) · VATEX-EVAL | Kendall tau_b: 28.1 -> 27.1 | Hurt |
| replace BLIP with IDEFICS-3 as caption generator (ablation) | PAC-S++ (ViT-B/32) · VATEX-EVAL | Kendall tau_b: 28.1 -> 26.9 | Hurt |
| replace SDv1.5 with FLUX.1 as image generator (ablation) | PAC-S++ (ViT-B/32) · Flickr8k-Expert | Kendall tau_c: 54.5 -> 54.8 | No clear effect |
Closest papers by meaning
- CAPability: A Comprehensive Visual Caption Benchmark for Evaluating Both Correctness and Thoroughness (NeurIPS 2025 · similarity 0.88)
- CaptionQA: Is Your Caption as Useful as the Image Itself? (CVPR 2026 · similarity 0.87)
- G-VEval: A Versatile Metric for Evaluating Image and Video Captions Using GPT-4o (AAAI 2025 · similarity 0.87)
- CapRL: Stimulating Dense Image Caption Capabilities via Reinforcement Learning (ICLR 2026 · similarity 0.87)
- Mining Fine-Grained Image-Text Alignment for Zero-Shot Captioning via Text-Only Training (AAAI 2024 · similarity 0.87)
- A Picture is Worth More Than 77 Text Tokens: Evaluating CLIP-Style Models on Dense Captions (CVPR 2024 · 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.
