MediaPipe Pose in Computer Vision - Model Pipeline Trace

Model Pipeline - MediaPipe Pose

MediaPipe Pose is a machine learning pipeline that detects human body landmarks in images or videos. It tracks 33 key points on the body to understand pose and movement in real time.

Data Flow - 5 Stages

1Input Image

1 frame x 1920 x 1080 pixels x 3 color channels→Capture or load a color image frame→1 frame x 1920 x 1080 pixels x 3 color channels

A photo of a person standing in front of a plain background

↓

2Preprocessing

1 frame x 1920 x 1080 x 3→Resize and normalize image pixels to model input size→1 frame x 256 x 256 x 3

Image resized to 256x256 pixels with pixel values scaled between 0 and 1

↓

3Pose Landmark Detection Model

1 frame x 256 x 256 x 3→Run neural network to predict 33 body landmarks→1 frame x 33 landmarks x 3 coordinates (x, y, visibility)

Coordinates like (0.45, 0.60, 0.98) for right shoulder landmark

↓

4Postprocessing

1 frame x 33 landmarks x 3→Map normalized landmark coordinates back to original image size→1 frame x 33 landmarks x 3 (pixel x, pixel y, visibility)

Right shoulder at pixel (864, 648) with visibility 0.98

↓

5Output Visualization

1 frame x 33 landmarks x 3→Draw landmarks and connections on original image→1 frame x 1920 x 1080 x 3 with overlay

Image showing dots and lines over the person's body joints

Training Trace - Epoch by Epoch

Loss
2.5 |*       
2.0 | *      
1.5 |  *     
1.0 |   *    
0.5 |    *   
0.0 |     *  
     --------
     Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	2.5	0.30	Initial training with high loss and low accuracy
5	1.2	0.55	Loss decreased, accuracy improving as model learns body shapes
10	0.7	0.75	Model captures pose landmarks more accurately
15	0.4	0.85	Good convergence, landmarks detected reliably
20	0.25	0.92	Final epoch with low loss and high accuracy

Prediction Trace - 4 Layers

Layer 1: Input Image

Layer 2: Neural Network Forward Pass

Layer 3: Coordinate Mapping

Layer 4: Visualization Overlay

Model Quiz - 3 Questions

What is the shape of the model's output after landmark detection?

A 1920 x 1080 x 3 image

B 256 x 256 x 3 image tensor

C 33 landmarks x 3 coordinates

D 1 landmark x 2 coordinates

Key Insight

MediaPipe Pose uses a neural network to detect 33 body landmarks by resizing input images and predicting normalized coordinates. Training reduces loss and improves accuracy, enabling real-time pose estimation with high confidence.