WEAKLY SUPERVISED FOOD IMAGE SEGMENTATION USING CLASS ACTIVATION MAPS.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Nutrition Informatics

Background:

• Food image segmentation is vital for dietary assessment but requires extensive pixel-level data.
• Current methods are hindered by the lack of available and costly food image datasets.
• Weakly supervised learning offers a potential solution to data scarcity.

Purpose of the Study:

• To develop a food image segmentation method using only image-level annotations.
• To address the challenge of limited pixel-level labeled data in food image analysis.
• To improve the feasibility of automated dietary assessment systems.

Main Methods:

• A weakly supervised convolutional neural network (CNN) was designed.
• A graph-based segmentation approach utilizing class activation maps was proposed.
• The model was trained on food datasets using image-level labels.

Main Results:

• The proposed method achieved competitive accuracy in food image classification.
• The approach demonstrated effectiveness in segmentation tasks.
• It successfully leveraged image-level annotations for pixel-level segmentation.

Conclusions:

• Weakly supervised learning is a viable approach for food image segmentation.
• The developed method reduces the dependency on expensive pixel-level annotations.
• This technique can advance image-based dietary assessment and management systems.