Multi-modal classification: specialized captioning

[herongrove]

Train a custom caption generator specifically for pictures of food.

• [x] Download BigOven recipes and images
• [x] Use 350K15K BigOven and 350K55K randomlycarefully selected ImageNet images to make a food-or-not classifier
• [x] Use food-or-not classifier to make % food pictures feature for SVM classifier
• [ ] Use BigOven images and recipes to train a caption generator
• [ ] Evaluate caption generator -- unclear whether this kind of captioning is possible
• [ ] Evaluate ingredient-finding with respect to off-the-shelf food image segmenting software
• [ ] Generate SVM features from captions, e.g. foodCap:2, foodCap:cups, foodCap:flour