IngredSAM: Open-World Food Ingredient Segmentation via a Single Image Prompt
Author Information
Author(s): Chen Leyi, Wang Bowen, Zhang Jiaxin, Cherifi Hocine
Primary Institution: College of Food Science and Nutritional Engineering, China Agricultural University
Hypothesis
Can a one-shot, open-world segmentation framework effectively segment food ingredients without requiring extensive training?
Conclusion
IngredSAM outperforms existing methods in food ingredient segmentation, achieving significant improvements in accuracy on standard datasets.
Supporting Evidence
- IngredSAM achieved 2.85% and 6.01% better performance than previous state-of-the-art methods on FoodSeg103 and UECFoodPix datasets.
- The framework allows for train-free segmentation of any food ingredient.
- IngredSAM demonstrated superior performance compared to various supervised and open-world segmentation methods.
Takeaway
IngredSAM is a new tool that helps computers recognize different food ingredients in pictures without needing a lot of training.
Methodology
The study introduces IngredSAM, which uses visual foundation models and prompt engineering to segment food ingredients from images without training.
Potential Biases
Potential biases in dietary data collection could skew nutritional recommendations.
Limitations
The model relies on high-quality images for effective segmentation and may struggle in environments lacking such images.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website