Skip to content

API.md

Latest commit

 

History

History
65 lines (46 loc) · 1.72 KB

API.md

File metadata and controls

65 lines (46 loc) · 1.72 KB

API Reference

This project exposes a direct prediction API alongside the browser UI.

Request-Response Flow

sequenceDiagram
	participant C as Client
	participant A as Flask /api/predict
	participant P as process_frame_data
	participant MP as MediaPipe
	participant ML as .pkl Model

	C->>A: POST image payload
	A->>P: validate + decode
	P->>MP: detect landmarks
	MP-->>P: feature points
	P->>ML: predict class
	ML-->>P: class index
	P-->>A: prediction + metrics
	A-->>C: JSON response
Loading

Element Guide

  • Client: Caller script, service, or frontend sending image input.
  • Flask /api/predict: API route that validates payload and formats output.
  • process_frame_data: Shared inference logic used by UI and API.
  • MediaPipe: Hand-landmark detection stage.
  • .pkl Model: Trained classifier artifact used for final class prediction.

Predict

POST /api/predict

Use this endpoint when you want to send a single image and get a prediction back without using the webpage.

Supported Inputs

  • multipart/form-data with file field image
  • application/json with base64 content in image_base64

Optional Fields

  • show_landmarks: true or false
  • include_visuals: true or false

Response Fields

  • prediction: predicted class label
  • brightness: frame brightness score
  • contrast: frame contrast score
  • low_brightness: whether the image is below the configured threshold
  • model: active model filename

When include_visuals is enabled, the response also includes:

  • processed_frame
  • original_frame when landmarks are hidden

Notes

  • The endpoint reuses the same inference pipeline as the web UI.
  • The webpage stays available at the same time, so both flows can be used together.