Facial Landmark Detection Model

Title: Facial Landmark detection on edge devices
Target Model: ONNX format for maximum portability.
Input for the Model: Face Bounding box
Output: The model should output an estimate of 478 3-dimensional face landmarks.
Project Output: Training Code, Inference Code in Python, list of datasets used
Hardware Optimization:
● Primarily target NVIDIA Jetson Nano for edge deployment.
● Minimise CPU and RAM usage for efficient resource utilisation.
● Consider optimisations for both CPU and GPU (NVIDIA CUDA) execution.
Use Case: The above model will be used to calculate the head pose and also the emotions of
the face
Title: Custom YOLO object detector for edge devices
Target Model: ONNX format for maximum portability.
Input for the Model: Frame
Output: The model should detect Person, Smart Phone, Cigarette, Bicycle, Car,
Motorcycle, Bus, Truck, Indian auto rickshaw, Traffic lights, Traffic Sign Boards, Cat, Dog,
Sheep, Cow, Bottle.
Project Output: Training Code, Inference Code in Python, list of datasets used
Hardware Optimization:
● Primarily target NVIDIA Jetson Nano for edge deployment.
● Minimise CPU and RAM usage for efficient resource utilisation.
● Consider optimisations for both CPU and GPU (NVIDIA CUDA) execution.
Use Case: The above model will be used to detect objects in the frame.
Title: Lane detection
Target Model: ONNX format for maximum portability.
Input for the Model: Frame
Output: Detected Lane Coordinates
Project Output: Training Code, Inference Code in Python, list of datasets used
Hardware Optimization:
● Primarily target NVIDIA Jetson Nano for edge deployment.
● Minimise CPU and RAM usage for efficient resource utilisation.
● Consider optimisations for both CPU and GPU (NVIDIA CUDA) execution.
Use Case: The above model will be used to detect lanes on the road for a Road Safety system