SOPHON ACADEMY / Task
Introduction

Introduction

  • Objective

Develop an emotion analysis tool based on facial image classification that can identify and evaluate emotional states (such as happiness, sadness, anger, surprise, etc.) by analyzing users' facial expressions. This tool will leverage the NPU computing power of the Milk-V Duo to achieve efficient face detection, feature extraction, and emotion recognition, applicable to fields such as social interaction, mental health monitoring, and customer service.

 

  • Detailed Task Description

1. Face Detection and Feature Extraction:
a. Implement a real-time face detection module that can accurately identify faces from video captured by a camera.
b. Extract features from detected faces, including the localization of facial key points (such as eyes, eyebrows, mouth, etc.).

 

2. Emotion Analysis Model Development:
a. Deploy a pre-trained emotion analysis model that uses the NPU power of the Milk-V Duo to recognize emotional states based on facial features.
b. The model should handle different lighting conditions and facial obstructions to ensure the accuracy of emotion recognition.

 

3. Results Display and Feedback:
a. Design an intuitive results display interface that shows the recognized emotional states and corresponding emotional intensity scores in real time.
b. Provide a user feedback mechanism that allows users to correct analysis results to further train and optimize the model.

 

  • Performance Requirements
  1. The total response time for face detection and emotion analysis should not exceed 2 seconds.
  2. The accuracy of emotion recognition should be no less than 90%, with emotional intensity scores highly correlated to the user's actual emotional state.
  3. Memory Usage: Optimize memory allocation to ensure the efficiency of face detection and emotion analysis model operation while not exceeding the 256MB memory limit.

 

  • Acceptance Criteria
  1. The face detection module should run stably on the Milk-V Duo, accurately identifying faces in the video.
  2. The emotion analysis model should accurately recognize a variety of emotional states and provide corresponding emotional intensity scores.
  3. The results display interface should be clear, updating analysis results in real time, with an effective user feedback mechanism.
  4. Functionality Testing: The system should pass tests with at least 1000 facial images of different emotional states to ensure the accuracy and stability of the analysis results.
  5. The submitted project should include complete source code, documentation, and necessary resource files for subsequent maintenance and optimization.

 

By completing this task, developers will be able to showcase the potential application of the Milk-V Duo in the fields of computer vision and emotion analysis, providing users with an innovative tool to help them better understand and analyze human emotional states. This will not only enhance user experience but also provide developers with practical experience in facial emotion analysis technology.