SKH

Hi Robin, It seems that you have fixed the issue regarding training with multiple GPUs. However, please notice that the current implementation uses 'dataparallel' to wrap the model. The problem...

Hi Robin, `_generate_matrix` is not causing the problem. Two other places in the `acuracy_metric` module are causing the issue for your usecase. One is `plot_confusion_matrix` function which explicitly removes class...

Hi Robin, This is the exact dataset. "chips_filtered_13_classes_complete" was our internal name for the dataset. Please update the config with the new name.

Hi Robin, That function is from an older release of the code which we decided to remove later on. I will add the correct one sometime this week.

You can use this code snippet to visualize random samples from your custom dataset. Notice that you call the function before passing the dataset into the pytorch dataloader.

import numbers import matplotlib.pyplot as plt import random import torch def show_random_patches(dataset, sample_num, rgb_bands=(3, 2, 1)): """ Plots a user-defined number of image chips and the corresponding labels. """ if...

As the model is big I decided to use dropout to overcome overfitting I checked 0.1, 0.15, 0.20 and 0.25 and decided to use 0.15. If you want to optimize...