Model Training
PLAINTEXT
Snippet Details and Actions
2.2 Model Training
An environment is developed using deep learning algorithms in python 3.6 with Tensorflow-gpu 1.15.5 and keras-appplications 1.0.8 to train and test the algorithm. Part of the code is downloaded from (https://github.com/Jianningli/autoimplant), however, the algorithm is designed according to our own requirements using separate encoder-decoder network and is tested on Lenovo Legion Ryzen 7 with NVIDIA GTX 1660Ti (6GB GPU Memory).
This AI based algorithm was trained many times using 20,000 epochs to fully understand the defected areas of the skull. During the training, model algorithm detects the defected area and boundary layer, identify the shape, size and geometry of the defect of various skulls added in the training. In the second step, it predicts the required shaped as detected from the defected skull to fit it on the required defect. The training and prediction of the skull from defected to fine be shown in Figure 2.