Experiment notebooks used to get a better performance or visualization in the final training template.
Notebook to display statistics and description about model architectures.
Template notebook using CPU, which takes approximately 12 hours.
Template notebook using GPU, which takes approximately 1 hour.
Template notebook using GPU and techniques, experimenting the RAM usage and time to train.
Template notebook using TPU, experimenting the RAM usage and time to train which takes approximately less than 30 minutes, it only works with few images.
Attempt to use TPU with entire dataset. It requires a reasonable quantity of RAM. Also TCU have limitations about validation sets and other configurations which GPU provides.
Notebook to test the possibility of downloading the dataset instead of transfering from Google Drive.
Notebook to experiment the metrics and visualization of metrics as model output.
Notebook to explore segmentation possibilities.
Notebook to evaluate the time of string concatenation, achieving the lower time to import the images into memory.
Both experiments to test if the template works as expected.
Experiments with Generative Adversarial Networks for generating only melanoma images.
Testing Gans generation.
Testing with batch size of 1.
Testing with batch size of 5.
Notebook for generate images from the chosen models.
Template notebook for training with GANs