model and weights where to find?

[jinjunfei]

After training the network, where are the results saved? What model to choose for testing? What are weights?

[marinbenc]

Hi, you can find the trained weights in the releases section here:

https://github.com/marinbenc/medical-polar-training/releases/tag/v1.0

While training, by default the checkpoints are saved inside logs/ with a folder named after the current date, however you can set this with the --logs parameter.

You can choose a trained checkpoint and an appropriate model type for that checkpoint for testing.

I hope this helps. let me know if you have any other questions.

[marinbenc]

From our experiments, the best models were when we combined a stacked hourglass model (with augmentation) with a polar U-Net with agumentation. In your case, that would mean using polyp_stacked_hourglass_sigma_8 to find centerpoints, and then use those centerpoints to transform the input image to polar coordinates, and then give that image to polyp_unet_polar_aug to perform the segmentation.

You can see this approach in test_centerpoint_model.py.

I will direct you to our paper for more details.

For all networks, we use the Adam optimizer with a learning rate of 10e-3. A batch size of 8 was used for all networks except the center-point model, where a batch size of 6 was used for the lesion and liver datasets, and 8 for all remaining datasets. We trained all models up to a maximum of 200 epochs and used checkpoints after each epoch to store the model with the best validation loss.

[jinjunfei]

Thank you very much!