Task Description

The participants need to develop a model that will take an Indian thali image as input and detect each food item present in that image.

A typical Indian Thali contains a variety of dishes served together on a single platter. For example, white rice served with yellow dal and salad can be a typical Indian Thali. However, the food items are not fixed.

Input - The model will take an RGB image of Indian Thali (of any resolution) as input.

Output - The model should generate the following outputs.

  a) bounding box containing each food item in a particular thali.

b) labels for each bounding box that classifies the surrounded food item.

c) Calorie estimation for the whole thali (Optional)

Basic Instructions - Candidates are free to use any tools and technologies they want to. Some of them are mentioned below.

• Pytorch, TensorFlow, Keras, etc.

• YOLO, ViT, SAM etc.

• Pretrained or fine-tuned, CLIP based models for zero-shot learning, etc.

Evaluation Metrics - The model performance will be measured using the following evaluation metrics.

• mAP (mean average precision) - for bounding box prediction and classification