Human Body Part Identification Model

This is a computer vision model designed to identify 17 parts of the human body. The parts are:

• bm_arm_left
• bm_arm_right
• bm_body_up
• bm_hand_left
• bm_hand_right
• bm_head
• bm_leftfoot
• bm_leftknee
• bm_leftleg
• bm_leftsleg
• bm_neck
• bm_rightfoot
• bm_rightknee
• bm_rightleg
• bm_rightsleg
• bm_sarm_left
• bm_sarm_right

Dataset

• Find the dataset here.
• Roboflow offers various image datasets for deep learning, including object detection, classification, semantic segmentation, instance segmentation, and keypoint detection datasets.

Purpose of the Model

• Emergency Response: The model aims to assist emergency response units by identifying body parts to detect injuries.
• Injury Detection: Plans to combine this dataset with an 'Injury Dataset' to detect and identify the type and location of injuries on a person.

Training and Testing

• Current Status (As of 17th May 2024):
  • Trained using YOLOv8 and Detectron2.
  • Initial training and tests with YOLOv8 showed low accuracy (e.g., bm_body_up was identified as bm_head).
  • Low accuracy may be due to:
    • Limited dataset (318 images).
    • Low number of epochs (100).

-Find the weights of the two models under the yolov8_model_output/100 epochs and detectron_model_output folders. -I have also included the jpg outputs of the models after training them and testing them on a sample image of myself.

• Future Plans:
  • Augment dataset.
  • Increase the number of epochs used on YOLOv8 during training.
  • Re-test to improve accuracy.

Challenges

• Limited Data: Training on only 318 images.
• Computational Limits: Colab deactivated GPU during Detectron2 testing.

Thrilling!!!!!!!!!