Research map / Vision-language and multimodal / CLIP and zero-shot
No Hard Negatives Required: Concept Centric Learning Leads to Compositionality without Degrading Zero-shot Capabilities of Contrastive Models
Hai X. Pham, David T. Hoffmann, Ricardo Guerrero, Brais Martinez
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 |
|---|---|---|---|
| fine-tune on DreamLIP CC3M with standard loss only (ablation) | SigLIP ViT-B/16 · SugarCrepe average | accuracy: 79.5 -> 81.7 | Improved |
| add noun-phrase concept loss (npc) (ablation) | C2LIP · SugarCrepe average | accuracy: 81.7 -> 85.3 | Improved |
| add cross-attended concept-aware loss (xac) (ablation) | C2LIP · SugarCrepe average | accuracy: 85.3 -> 85.6 | Improved |
| add npc loss (ablation) | C2LIP · Flickr8K retrieval | Recall@5: 95.2 -> 96.3 | Improved |
| add xac loss (ablation) | C2LIP · SugarCrepe Replace attribute | accuracy: 89.7 -> 89.2 | Hurt |
| use C2LIP objectives instead of plain CC3M fine-tuning | C2LIP · ImageNet1K zero-shot | accuracy: 75.9 -> 73.5 | Hurt |
| use C2LIP objectives instead of plain CC3M fine-tuning | C2LIP · SugarCrepe++ Swap TOT | accuracy: 36.6 -> 44.2 | Improved |
| use C2LIP objectives instead of plain CC3M fine-tuning | C2LIP · Overall average | average score: 71.5 -> 75.0 | Improved |
| use C2LIP vision encoder in LLaVA | LLaVA · SugarCrepe average | accuracy: 83.5 -> 83.9 | Improved |
| use C2LIP vision encoder in LLaVA | LLaVA · SugarCrepe++ average | accuracy: 68.1 -> 69.7 | Improved |
| use lambda_npc 1, lambda_xac 0.01 instead of 0.5 and 0.5 (ablation) | C2LIP · SugarCrepe average | accuracy: 84.6 -> 85.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FG-CLIP — FG-CLIP: Fine-Grained Visual and Textual Alignment (ICML 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Contrasting Intra-Modal and Ranking Cross-Modal Hard Negatives to Enhance Visio-Linguistic Compositional Understanding (CVPR 2024 · similarity 0.90)
- Not Only Text: Exploring Compositionality of Visual Representations in Vision-Language Models (CVPR 2025 · similarity 0.89)
- TripletCLIP: Improving Compositional Reasoning of CLIP via Synthetic Vision-Language Negatives (NeurIPS 2024 · similarity 0.89)
- Enhancing Vision-Language Compositional Understanding with Multimodal Synthetic Data (CVPR 2025 · similarity 0.89)
- Progressive Compositionality in Text-to-Image Generative Models (ICLR 2025 · similarity 0.89)
- Learning Visual Composition through Improved Semantic Guidance (CVPR 2025 · similarity 0.89)
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.
