Research map / Vision-language and multimodal / Cross-modal learning
CHEF: Cross-modal Hierarchical Embeddings for Food Domain Retrieval
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use the hierarchical T+L+T recipe encoder rather than ACME | Recipe1M image-to-recipe retrieval · Recipe1M test, 1k | R@1: 45.4 -> 49.7 | Helped |
| Use the hierarchical T+L+L recipe encoder rather than ACME | Recipe1M image-to-recipe retrieval · Recipe1M test, 10k | R@1: 18.0 -> 20.9 | Helped |
| Use Tree-LSTM ingredient encoding instead of attention | Main ingredient prediction · Recipe1M test | R@1: 9.1 -> 47.0 | Helped |
| Remove the least-important ingredient and its occurrences in instructions (ablation) | T+T+L retrieval · Recipe1M test, image-to-recipe | R@1: 49.0 -> 51.1 | Helped |
| Remove the two least-important ingredients and their instruction occurrences (ablation) | T+T+L retrieval · Recipe1M test, image-to-recipe | R@1: 49.0 -> 0.5 | Hurt |
| Remove the three least-important ingredients and their instruction occurrences (ablation) | T+T+L retrieval · Recipe1M test, image-to-recipe | R@1: 49.0 -> 48.7 | Hurt |
| Remove the four least-important ingredients and their instruction occurrences (ablation) | T+T+L retrieval · Recipe1M test, image-to-recipe | R@1: 49.0 -> 46.3 | Hurt |
| Keep only the six highest-ranked ingredients (ablation) | T+T+L retrieval · Recipe1M test, image-to-recipe | R@1: 49.0 -> 44.7 | Hurt |
| Keep ingredients up to tree depth 4 (ablation) | T+T+L retrieval · Recipe1M test, image-to-recipe | R@1: 49.0 -> 49.7 | Helped |
| Use Tree-LSTM for sentence encoding and LSTM for instruction-level encoding instead of Tree-LSTM at both levels (ablation) | Recipe1M action-word extraction · Recipe1M test | verb percentage: 47.88 -> 77.07 | Helped |
| Replace the ingredient Set encoder with GRU attention in the LSTM-based retrieval configuration (ablation) | Recipe1M image-to-recipe retrieval · Recipe1M test, 1k | R@1: 48.0 -> 49.3 | Helped |
| Replace the ingredient Set encoder with Tree-LSTM in the LSTM-based retrieval configuration (ablation) | Recipe1M image-to-recipe retrieval · Recipe1M test, 1k | R@1: 48.0 -> 49.4 | Helped |
Closest papers by meaning
- Revamping Cross-Modal Recipe Retrieval With Hierarchical Transformers and Self-Supervised Learning (CVPR 2021 · similarity 0.95)
- Learning Cross-Modal Embeddings for Cooking Recipes and Food Images (CVPR 2017 · similarity 0.94)
- Pic2Prep: A Multimodal Conversational Agent for Cooking Assistance (AAAI 2025 · similarity 0.87)
- OVFoodSeg: Elevating Open-Vocabulary Food Image Segmentation via Image-Informed Textual Representation (CVPR 2024 · similarity 0.86)
- Retrieval Across Any Domains via Large-scale Pre-trained Model (ICML 2024 · similarity 0.84)
- DiCA: Disambiguated Contrastive Alignment for Cross-Modal Retrieval with Partial Labels (AAAI 2025 · similarity 0.84)
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.