Research map / Image generation / Autoregressive and latent generation
Inverse Cooking: Recipe Generation From Food Images
Check 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 |
|---|---|---|---|
| Generate instructions using both image and predicted ingredients instead of image alone | Inverse Cooking · Recipe1M test | perplexity: 9.66 -> 8.51 | Helped |
| Add ingredient conditioning to image-conditioned instruction generation (ablation) | I2R · Recipe1M test | perplexity: 9.66 -> 8.51 | Helped |
| Add image features to ingredient-conditioned instruction generation (ablation) | L2R · Recipe1M test | perplexity: 8.67 -> 8.51 | Helped |
| Attend over concatenated image and ingredient embeddings rather than independent attention (ablation) | Instruction transformer · Recipe1M validation | perplexity: 8.59 -> 8.50 | Helped |
| Attend over concatenated embeddings rather than sequentially, image first (ablation) | Instruction transformer · Recipe1M validation | perplexity: 8.53 -> 8.50 | Helped |
| Attend over concatenated embeddings rather than sequentially, ingredients first (ablation) | Instruction transformer · Recipe1M validation | perplexity: 8.61 -> 8.50 | Helped |
| Represent ingredients as an order-invariant set while modeling co-occurrences (ablation) | Ingredient prediction · Recipe1M validation | F1: 45.55 -> 48.26 | Helped |
| Randomly shuffle the ordered ingredient targets during training (ablation) | TF_list · Recipe1M validation | F1: 45.55 -> 43.58 | Hurt |
| Use target-distribution loss instead of binary cross-entropy (ablation) | Feed-forward ingredient predictor · Recipe1M validation | F1: 30.30 -> 44.11 | Helped |
| Model ingredient dependencies and order-invariant set structure rather than independent binary labels (ablation) | Ingredient prediction · Recipe1M validation | F1: 30.30 -> 48.26 | Helped |
| Use image-recipe retrieval instead of generated ingredient prediction | Recipe1M test ingredient prediction · Recipe1M test | IoU: 19.85 -> 32.11 | Hurt |
| Generate instructions rather than retrieve them | Recipe generation · Recipe1M test instructions | ingredient recall: $31.92 -> $75.47 | Helped |
Closest papers by meaning
- Learning Cross-Modal Embeddings for Cooking Recipes and Food Images (CVPR 2017 · similarity 0.89)
- CookGAN: Causality Based Text-to-Image Synthesis (CVPR 2020 · similarity 0.89)
- Revamping Cross-Modal Recipe Retrieval With Hierarchical Transformers and Self-Supervised Learning (CVPR 2021 · similarity 0.88)
- 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 0.84)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · similarity 0.82)
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.