Real-Time CNN Face Crop Integration
Budget: ₹1,500 – ₹12,500 INR
I have a face-recognition pipeline that needs to run end-to-end in the browser, crop the user’s face in real time, pass that crop to a freshly trained CNN, and immediately forward the model’s decision back to my PAI endpoint.
The backbone of the job is the CNN itself. I’ll supply mixed image and video datasets; your task is to design, train, and fine-tune a model that can perform confidently under live-camera conditions. Whether you prefer TensorFlow, Keras, PyTorch, or another deep-learning framework is up to you, as long as the final weights and inference code are portable.
On the front end, I need a lightweight web component—WebRTC or similar—that guides users to place their face inside an on-screen box. When the alignment is correct, the component should grab the frame, crop precisely around the face, and hand the image off to the model. The resulting prediction must be pushed straight to my PAI REST endpoint with minimal latency. Responsive behaviour across modern browsers is essential.
Deliverables
• Production-ready CNN model (training notebook / scripts, trained weights, inference wrapper)
• Browser-based camera module with face box alignment and cropping
• Integration bridge that streams the crop to the model and posts the response to PAI
• Clear setup instructions and a quick demo so I can reproduce the full flow locally
Time is critical; the sooner this pipeline is working, the better. If you have relevant examples or repos, that will help accelerate the decision.
The backbone of the job is the CNN itself. I’ll supply mixed image and video datasets; your task is to design, train, and fine-tune a model that can perform confidently under live-camera conditions. Whether you prefer TensorFlow, Keras, PyTorch, or another deep-learning framework is up to you, as long as the final weights and inference code are portable.
On the front end, I need a lightweight web component—WebRTC or similar—that guides users to place their face inside an on-screen box. When the alignment is correct, the component should grab the frame, crop precisely around the face, and hand the image off to the model. The resulting prediction must be pushed straight to my PAI REST endpoint with minimal latency. Responsive behaviour across modern browsers is essential.
Deliverables
• Production-ready CNN model (training notebook / scripts, trained weights, inference wrapper)
• Browser-based camera module with face box alignment and cropping
• Integration bridge that streams the crop to the model and posts the response to PAI
• Clear setup instructions and a quick demo so I can reproduce the full flow locally
Time is critical; the sooner this pipeline is working, the better. If you have relevant examples or repos, that will help accelerate the decision.