Research map / Vision-language and multimodal / Multimodal LLMs
BRIDGE: Bridging Gaps in Image Captioning Evaluation with Stronger Visual Cues
Sara Sarto, 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 |
|---|---|---|---|
| replace the transformer mapping module with a two-layer MLP (ablation) | BRIDGE · Flickr8k-Expert | Kendall tau_b: 53.1 -> 54.4 | Improved |
| add template captions as mapping module input (ablation) | BRIDGE · Flickr8k-Expert | Kendall tau_b: 53.7 -> 54.4 | Improved |
| add the regularization branch (ablation) | BRIDGE · Flickr8k-CF | Kendall tau_b: 35.7 -> 36.1 | Improved |
| add textual similarity with the pseudo-caption to CLIP-Score (ablation) | BRIDGE · Flickr8k-Expert | Kendall tau_b: 51.1 -> 54.4 | Improved |
| add visual similarity term to the score (ablation) | BRIDGE · Flickr8k-Expert | Kendall tau_b: 53.8 -> 54.4 | Improved |
| use 3 pseudo tokens per noun chunk instead of 1, 2, 4 or 8 (ablation) | BRIDGE · Flickr8k-Expert / CF / Pascal-50S | Kendall tau_b / accuracy: Expert 54.1 -> 54.4 (US=1 vs 3) | Mixed |
| add the mapping module on top of BLIP template captions (ablation) | BRIDGE · Flickr8k-Expert | Kendall tau_b: 48.6 -> 54.4 | Improved |
| use higher-quality template captions (BLIP-2 instead of Transformer captioner) (ablation) | BRIDGE · Flickr8k-Expert | Kendall tau_b: 54.0 -> 54.4 | No clear effect |
| weight contrastive loss by number of noun chunks (ablation) | BRIDGE · Flickr8k-CF | Kendall tau_b: 35.6 -> 36.1 | Improved |
| use grid-level visual features instead of global features (ablation) | BRIDGE · Flickr8k-CF | Kendall tau_b: 35.1 -> 36.1 | Improved |
| keep only mapping outputs at masked positions instead of full output (ablation) | BRIDGE · Flickr8k-Expert | Kendall tau_b: 52.0 -> 54.4 | Improved |
| replace CLIP-Score with BRIDGE for hallucination detection | CLIP ViT-B/32 metric · FOIL | accuracy: 87.2 -> 91.5 | Improved |
Closest papers by meaning
- CAPability: A Comprehensive Visual Caption Benchmark for Evaluating Both Correctness and Thoroughness (NeurIPS 2025 · similarity 0.90)
- CaptionQA: Is Your Caption as Useful as the Image Itself? (CVPR 2026 · similarity 0.89)
- A Picture is Worth More Than 77 Text Tokens: Evaluating CLIP-Style Models on Dense Captions (CVPR 2024 · similarity 0.88)
- Enhancing Descriptive Captions with Visual Attributes for Multimodal Perception (CVPR 2026 · similarity 0.88)
- G-VEval: A Versatile Metric for Evaluating Image and Video Captions Using GPT-4o (AAAI 2025 · similarity 0.88)
- Painting with Words: Elevating Detailed Image Captioning with Benchmark and Alignment Learning (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.
