FoodSeg103

FoodSeg103 is a new food image dataset containing 7,118 images. Images are annotated with 104 ingredient classes and each image has an average of 6 ingredient labels and pixel-wise masks. It's provided as a large-scale benchmark for food image segmentation.

Major Challenges:

1. High intra-variance of the same food ingredient with different cooking methods
2. Long-tail distribution
3. Complicated contexts

Image source: https://arxiv.org/pdf/2105.05409v1.pdf