Computer Vision for Posture

This project, titled “No Pain, Just Gain,” focused on improving human posture estimation using computer vision. We implemented and trained Google’s BlazePose model with TensorFlow, then experimented with architectural modifications to boost the Percentage of Correct Key-points (PCK) scores. By integrating Convolutional Block Attention Modules (CBAM), we enhanced the model’s intermediate feature representations, resulting in more accurate and robust pose estimation.