Research map / Label-efficient and robust learning / Recognition and long-tail
Ladder Loss for Coherent Visual-Semantic Embedding
Mo Zhou, Zhenxing Niu, Le Wang, Zhanning Gao, Qilin Zhang, Gang Hua
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 VSE++ ranking loss with the ladder loss | ResNet-152 visual-semantic embedding · MS-COCO (1000 Test Samples), image→sentence | CS@100: 0.238 -> 0.265 | Improved |
| Replace VSE++ ranking loss with the ladder loss | ResNet-152 visual-semantic embedding · MS-COCO (1000 Test Samples), image→sentence | R@1: 63.2 -> 66.7 | Improved |
| Replace VSE++ ranking loss with the ladder loss | ResNet-152 visual-semantic embedding · Flickr30K, image→sentence | CS@100: 0.126 -> 0.133 | Improved |
| Add a nonzero β₂ ladder term (β₂=0.25 rather than 0) (ablation) | CVSE++ (Res152) · MS-COCO, image→sentence | CS@100: 0.238 -> 0.265 | Improved |
| Increase β₂ from 0.25 to 1.0 (ablation) | CVSE++ (Res152) · MS-COCO, image→sentence | R@1: 66.7 -> 64.0 | Hurt |
| Increase ladder count from L=1 to L=2 (ablation) | CVSE++ (Res152) · MS-COCO, image→sentence | CS@100: 0.238 -> 0.265 | Improved |
| Increase ladder count from L=2 to L=3 (ablation) | CVSE++ (Res152) · MS-COCO, image→sentence | CS@200: 0.252 -> 0.274 | Improved |
| Increase ladder count from L=2 to L=3 (ablation) | CVSE++ (Res152) · MS-COCO, image→sentence | CS@100: 0.265 -> 0.267 | Improved |
| Increase ladder count beyond the chosen setting (the paper notes L=5) (ablation) | CVSE++ · MS-COCO | R@K | Hurt |
Closest papers by meaning
- NUDGE: Lightweight Non-Parametric Fine-Tuning of Embeddings for Retrieval (ICLR 2025 · similarity 0.87)
- QuARI: Query Adaptive Retrieval Improvement (NeurIPS 2025 · similarity 0.86)
- NV-Embed: Improved Techniques for Training LLMs as Generalist Embedding Models (ICLR 2025 · similarity 0.86)
- MERIT: Multilingual Semantic Retrieval with Interleaved Multi-Condition Query (NeurIPS 2025 · similarity 0.85)
- Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval (ICLR 2025 · similarity 0.85)
- Mitigating Semantic Collapse in Partially Relevant Video Retrieval (NeurIPS 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.
