<aside> 🛠

Binary Gender Classification for Pedestrians (91% Accuracy)

Analyzes images of people and classifies gender as male or female. Uses Microsoft BeIT (BERT Image Transformer) trained on PETA pedestrian attributes dataset. Works on photos, video frames, and security camera footage.

</aside>

What it does

Pedestrian Gender Recognition analyzes images containing people and predicts gender (male/female) with confidence scores. Binary classification model that outputs gender label with prediction confidence. Works on photographs, video frames, security camera footage, or any image containing people.

Problem it solves

• Real-time pedestrian gender categorization in security/monitoring systems
• Crowd composition analysis (male/female distribution in public spaces)
• Safety systems tracking worker presence by gender in specific zones
• Analytics for public space usage patterns
• Behavioral analysis systems

Input/Output

<aside>

• Input: Image containing one or more people

  

</aside>

<aside>

• Output: Gender classification with confidence

  • Label: "male" or "female"

  • Confidence score: 0.0–1.0

  • Example: {"gender": "male", "confidence": 0.94}

    

</aside>

Accuracy & Speed

• Performance:
  • Accuracy: 91% on PETA test dataset
  • Model: Binary gender classification
  • Training data: PETA (Pedestrian Attributes) dataset

Technical Details

| Architecture | • Base: Microsoft BeIT (BERT Image Transformer) • Pre-training: ImageNet supervised learning • Parameters: ~86M • Head: Binary classification layer (male/female) | | --- | --- | | Training | • Dataset: PETA (Pedestrian Attributes) dataset • Epochs: 5 • Batch size: 8 • Optimizer: Adam • Learning rate: 2e-05 • Labels: Binary (male/female |

Compliance & Provenance