Research map / Vision-language and multimodal / Cross-modal learning
Learning Structural Representations for Recipe Generation and Food Retrieval
Hao Wang, Guosheng Lin, Steven C. H. Hoi, Chunyan Miao
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 SGN tree structure to non-pretrained model | Non-pretrained recipe generator · Recipe1M recipe generation | BLEU: 7.23 -> 9.09 | Improved |
| add SGN tree structure to TIRG model | TIRG recipe generator · Recipe1M recipe generation | BLEU: 7.95 -> 9.24 | Improved |
| add SGN tree structure to VAL model | VAL recipe generator · Recipe1M recipe generation | BLEU: 8.83 -> 10.43 | Improved |
| add SGN tree structure to pretrained BERT model | Pretrained recipe generator · Recipe1M recipe generation | BLEU: 9.29 -> 12.75 | Improved |
| add SGN to pretrained model | Pretrained recipe generator · Recipe1M recipe generation | Perplexity: 7.52 -> 6.67 | Improved |
| add ingredient features (ablation) | Pretrained recipe generator · Recipe1M recipe generation | BLEU: 3.72 -> 9.29 | Improved |
| add image features (ablation) | Pretrained recipe generator · Recipe1M recipe generation | BLEU: 5.74 -> 9.29 | Improved |
| add SGN to improve generated recipe length | Pretrained recipe generator · Recipe1M recipe generation | Recipe average length: 66.9 -> 112.5 | Improved |
| add tree structure features to retrieval model (ablation) | Retrieval model with self-attention LSTM · Recipe1M image-to-recipe retrieval (1k test) | R@1: 52.5 -> 53.5 | Improved |
| add tree structure features to transformer recipe encoder (ablation) | Retrieval model (transformer) · Recipe1M image-to-recipe retrieval (1k test) | R@1: 53.4 -> 54.3 | Improved |
| use sentence plus depth node features instead of adjacency matrix projection (ablation) | Tree feature GAT · Recipe1M image-to-recipe retrieval (1k test) | R@1: 52.7 -> 53.5 | Improved |
| replace concatenation with TIRG fusion of tree and text features (ablation) | Retrieval model (transformer) · Recipe1M image-to-recipe retrieval (1k test) | R@1: 54.3 -> 54.7 | Improved |
Closest papers by meaning
- CHEF: Cross-modal Hierarchical Embeddings for Food Domain Retrieval (AAAI 2021 · similarity 0.94)
- Structure-Aware Generation Network for Recipe Generation from Images (ECCV 2020 · similarity 0.92)
- Learning Cross-Modal Embeddings for Cooking Recipes and Food Images (CVPR 2017 · similarity 0.91)
- OVFoodSeg: Elevating Open-Vocabulary Food Image Segmentation via Image-Informed Textual Representation (CVPR 2024 · similarity 0.87)
- Pic2Prep: A Multimodal Conversational Agent for Cooking Assistance (AAAI 2025 · similarity 0.87)
- Compositional Entailment Learning for Hyperbolic Vision-Language Models (ICLR 2025 · similarity 0.83)
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.
