Real-Time Windows Face Recognition System
Budget: $250 – $750 USD
I need an end-to-end, live face-recognition model that runs smoothly on Windows and authenticates users from a webcam feed in real time.
The pipeline must follow the architecture I already have in mind:
• Feature extraction: implement Global Search ShuffleNet coupled with a Generative Adversarial Network (GSS-GAN) from scratch or by extending public research code.
• Face cognition / matching: build the classifier with a Convolutional Neural Network optimised for low latency.
The model should open a webcam stream, detect a face, apply GSS-GAN for robust feature vectors, and pass them through the CNN to decide whether the face belongs to an enrolled user. An accuracy benchmark on a small hold-out set is fine for now, but the live demo has to stay above 25 fps on a mid-range Windows laptop.
Deliverables
- Clean, well-commented Python code (TensorFlow or PyTorch) with requirements.txt
- Trained weights for both GSS-GAN and CNN
- A lightweight Windows app or script that launches the webcam, performs the full pipeline, and prints “Authenticated / Rejected” on screen
- Short README explaining setup, training procedure, and how to add new users
Acceptance criteria
1. End-to-end demo runs locally with no missing dependencies.
2. Identification accuracy ≥ 95 % on the sample set I will provide.
3. Average processing speed ≥ 25 fps on live video.
If you have hands-on experience with ShuffleNet, GANs, and real-time OpenCV pipelines, this should be a straightforward build. Let me know about any prior work with similar CNN-based authentication systems and how long you’ll need to deliver the first prototype.
The pipeline must follow the architecture I already have in mind:
• Feature extraction: implement Global Search ShuffleNet coupled with a Generative Adversarial Network (GSS-GAN) from scratch or by extending public research code.
• Face cognition / matching: build the classifier with a Convolutional Neural Network optimised for low latency.
The model should open a webcam stream, detect a face, apply GSS-GAN for robust feature vectors, and pass them through the CNN to decide whether the face belongs to an enrolled user. An accuracy benchmark on a small hold-out set is fine for now, but the live demo has to stay above 25 fps on a mid-range Windows laptop.
Deliverables
- Clean, well-commented Python code (TensorFlow or PyTorch) with requirements.txt
- Trained weights for both GSS-GAN and CNN
- A lightweight Windows app or script that launches the webcam, performs the full pipeline, and prints “Authenticated / Rejected” on screen
- Short README explaining setup, training procedure, and how to add new users
Acceptance criteria
1. End-to-end demo runs locally with no missing dependencies.
2. Identification accuracy ≥ 95 % on the sample set I will provide.
3. Average processing speed ≥ 25 fps on live video.
If you have hands-on experience with ShuffleNet, GANs, and real-time OpenCV pipelines, this should be a straightforward build. Let me know about any prior work with similar CNN-based authentication systems and how long you’ll need to deliver the first prototype.