sababishraq/foodsense-dataset

FoodSense Dataset

Human-annotated food images with sensory ratings (taste, smell, texture, sound on a 1-5 scale) and free-text descriptors, from the FoodSense project.

Project Page | Model | Code

Dataset summary

Statistic	Value
Total food images	2,987
Participants	8,382
Total annotations	66,842 participant-image pairs
Mean annotators per image	22.38 (SD 2.02)
Sensory dimensions	4 (taste, smell, texture, sound)
Rating scale	1-5 (rescaled from original 1-7 Likert)
Training subset (CanInfer filtered)	2,915 images / 58,443 annotations

Source images are drawn from the Yelp Open Dataset.

Data layout

• Images: JPEG files named NNNN_<yelp_photo_id>.jpg at the repository root, where NNNN is a zero-padded index.
• metadata.csv: Sits alongside the images at root. Each row has a file_name column matching the image filename, plus aggregated ratings and example descriptors.

This layout follows the Hub ImageFolder convention so the Dataset Viewer can resolve images and columns automatically.

Sample images

Index	file_name	Preview
1	0001_01lamiW2bWW0rXlllNHYMA.jpg
2	0002_01zZeZBIFZ82S5XmA4GYJg.jpg
3	0003_05KUDlEPkMLF-fTrTk4qxQ.jpg

Columns in metadata.csv

Column	Description
file_name	Image filename (matches files in the repo root)
image_id	Annotator / set identifier
taste, smell, texture, sound	Mean ratings across annotators (1-5)
taste_desc, smell_desc, texture_desc, sound_desc	Example free-text descriptions

Citation

@inproceedings{ishraq2026foodsense,
  title     = {FoodSense: A Multisensory Food Dataset and Benchmark for
               Predicting Taste, Smell, Texture, and Sound from Images},
  author    = {Ishraq, Sabab and Aarushi, Aarushi and Jiang, Juncai and Chen, Chen},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and
               Pattern Recognition (CVPR)},
  year      = {2026}
}

License

Creative Commons Attribution 4.0 International (cc-by-4.0).