This project aims to classify Face Skin Diseases by using the power of deep learning concepts.
Training Classification Models with Intel Optimized TensorFlow on Intel DevCloud. Then model inference on NVIDIA Jetson Nano in ONNX and TensorRT formats.
Firstly, custom CNN Models were used and then pre-trained MobileNet V2 was used with Transfer Learning. At the end, Fine tuning was done to increase the model performance. You can see how the accuracy on unused test images increases in each step.
The model_training notebook can be taken as pipeline and used in different platforms. But it was actually designed to train the models with Intel Extension for TensorFlow on Intel Developer Cloud to accelerate training.
Later, the trained TensorFlow models are converted into ONNX and TensorRT formats and used in NVIDIA Jetson Nano to inference.