RecipeQA: A Dataset for Multimodal Comprehension of Cooking Recipes
Recipe1M+: A Dataset for Learning Cross-Modal Embeddings for Cooking Recipes and Food Images
RecipeNLG dataset contains 2.2 million recipes; see the corresponding paper
Epicurious Food Ingredients and Recipes Dataset with Images; 13.5K recipes
YouCook2 is one of the largest task-oriented, instructional video dataset in the vision community
Food.com Crawled data from Food.com (GeniusKitchen) online recipe aggregator, 280K+ recipes
Food related datasets at data.world
The Rakuten dataset with about 800,000 recipes and images (Japanese); see also the paper describing this dataset
MIRecipe dataset consists of 239,973 instructional images for 26,725 (Chinese) recipes, including recipe name, recipe ID, brief descriptions, ingredients, seasonings, instructions (both text and image) and tips
Carnegie Mellon Recipe Database contains 260 annotated recipes; see also the corresponding paper which also introduces the Minimal Instruction Language for the Kitchen (MILK).

Representing recipes

Google provides a useful schema for representing recipes: Recipe Schema Markup | Google Search Central | Documentation | Google for Developers. It includes, amongst others, the following elements:

name of the dish
cooking time, i.e. the time it takes to actually cook the dish (in ISO 8601 format)
preparation time, i.e. the time it takes to prepare ingredients and workspace for the dish
ingredients used in the recipe
recipe instructions, i.e. the steps to make the dish
yield, i.e. the quantity produced by the recipe, either the number of servings produced or a different unit (for example, number of items)

Recipe selection:

category, i.e. the type of meal or course your recipe is about, e.g., dinner, main course, dessert, snack
cuisine, i.e. the region associated with your recipe, e.g., French, Mediterranean, or American

Visual support:

Additional info:

Recipes