ESP32 Real-Time Oral Cancer Detector -- 2
Budget: ₹12,500 – ₹37,500 INR
I’m building a proof-of-concept that lets an ultra-low-power board—specifically an ESP32—capture image data and immediately flag early-stage oral cancer. Because the hardware budget is tiny, I need a novel, highly compressed AI architecture that still reaches clinically useful accuracy and inference speed at the edge.
The device will ingest live image data from a camera module, run on-device preprocessing, then execute an AI-based model (think pattern-recognition and other machine-learning techniques rather than simple thresholding) fast enough to give instant feedback without off-loading to the cloud. Latency, memory footprint, and power draw must all stay within the ESP32’s limits.
If you’re interested, send a detailed project proposal outlining:
• The model architecture you would adapt or design (e.g., TinyML CNN, MobileNet variants, quantization/pruning strategy)
• How you will tackle data collection, augmentation, and on-chip preprocessing
• Your plan for optimizing inference time and RAM/flash usage on the ESP32 toolchain (ESP-IDF / Arduino, TensorFlow Lite Micro, or similar)
• A validation strategy showing accuracy, sensitivity, and false-positive rates on a held-out image set
Final deliverables include compiled firmware, source code, trained weights, and concise build/run documentation so I can reproduce results on my own ESP32 board.
The device will ingest live image data from a camera module, run on-device preprocessing, then execute an AI-based model (think pattern-recognition and other machine-learning techniques rather than simple thresholding) fast enough to give instant feedback without off-loading to the cloud. Latency, memory footprint, and power draw must all stay within the ESP32’s limits.
If you’re interested, send a detailed project proposal outlining:
• The model architecture you would adapt or design (e.g., TinyML CNN, MobileNet variants, quantization/pruning strategy)
• How you will tackle data collection, augmentation, and on-chip preprocessing
• Your plan for optimizing inference time and RAM/flash usage on the ESP32 toolchain (ESP-IDF / Arduino, TensorFlow Lite Micro, or similar)
• A validation strategy showing accuracy, sensitivity, and false-positive rates on a held-out image set
Final deliverables include compiled firmware, source code, trained weights, and concise build/run documentation so I can reproduce results on my own ESP32 board.