Research map / Label-efficient and robust learning / Recognition and long-tail
LCEval: Learned Composite Metric for Caption Evaluation
Naeha Sharif, Lyndon White, Mohammed Bennamoun, Wei Liu, Syed Afaq Ali Shah
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 learning (one-layer NN) over equal-weight feature combination (ablation) | LCEval · COMPOSITE (caption-level) | Kendall tau: 0.273 -> 0.375 | Improved |
| use two-layer NN instead of one-layer (ablation) | LCEval · COMPOSITE (caption-level) | Kendall tau: 0.375 -> 0.418 | Improved |
| add a second hidden layer (three-layer NN) (ablation) | LCEval · COMPOSITE (caption-level) | Kendall tau: 0.418 -> 0.417 | No clear effect |
| learn SPIDEr combination with network (ablation) | SPIDEr · COMPOSITE (caption-level) | Kendall tau: 0.357 -> 0.367 | Improved |
| use FastText instead of GloVe for MOWE (ablation) | LCEval · COMPOSITE (caption-level) | Kendall tau: 0.414 -> 0.418 | Improved |
| add lexical features (ablation) | LCEval · COMPOSITE (caption-level) | Kendall tau: 0.366 -> 0.418 | Improved |
| add semantic features (ablation) | LCEval · COMPOSITE (caption-level) | Kendall tau: 0.387 -> 0.418 | Improved |
| add syntactic features (ablation) | LCEval · COMPOSITE (caption-level) | Kendall tau: 0.413 -> 0.418 | Improved |
| train on all three captioning models instead of Adaptive Attention only (ablation) | LCEval · COMPOSITE (caption-level) | Kendall tau: 0.403 -> 0.418 | Improved |
| train on 132,984 examples instead of 88,656 from Show and Tell only (ablation) | LCEval · COMPOSITE (caption-level) | Kendall tau: 0.421 -> 0.418 | No clear effect |
| add more machine caption variety to training (ablation) | LCEval · PASCAL50S HHC | accuracy: 58.7 -> 64.0 | Improved |
| increase number of reference captions from 1 to 12 (ablation) | LCEval · PASCAL50S | accuracy: 0.740 -> 0.804 | Improved |
Closest papers by meaning
- Polos: Multimodal Metric Learning from Human Feedback for Image Captioning (CVPR 2024 · similarity 0.88)
- CaptionQA: Is Your Caption as Useful as the Image Itself? (CVPR 2026 · similarity 0.86)
- G-VEval: A Versatile Metric for Evaluating Image and Video Captions Using GPT-4o (AAAI 2025 · similarity 0.86)
- CAPability: A Comprehensive Visual Caption Benchmark for Evaluating Both Correctness and Thoroughness (NeurIPS 2025 · similarity 0.86)
- Painting with Words: Elevating Detailed Image Captioning with Benchmark and Alignment Learning (ICLR 2025 · similarity 0.85)
- Revisiting text-to-image evaluation with Gecko: on metrics, prompts, and human rating (ICLR 2025 · similarity 0.85)
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.
