Some question about data preprocessing.

[chenslcool]

This project is a great framework. But I have some question to ask.

1. I notice that in preprocessing phase, you crop each image into a box (larger than patch size(112, 112, 80)) according to label mask and get a LA region. But later, in Dataset LAHeart, you add a random crop transform to crop image to (112, 112, 80) again. Why this? I think just croping to (112, 112, 80) and remove the random crop in transform is enough. Random crop may lose some LA information.
2. In the testing phase, why do you make several inference on patches of one image? If croping to (112, 112, 80) in preprocessing , I think it will be easier to make test with no need for patching. Besides, I think this patching and average method used in testing may act like "ensemble learning", which will improve performance.

[Allenem]

This project is a great framework. But I have some question to ask.

1. I notice that in preprocessing phase, you crop each image into a box (larger than patch size(112, 112, 80)) according to label mask and get a LA region. But later, in Dataset LAHeart, you add a random crop transform to crop image to (112, 112, 80) again. Why this? I think just croping to (112, 112, 80) and remove the random crop in transform is enough. Random crop may lose some LA information.
2. In the testing phase, why do you make several inference on patches of one image? If croping to (112, 112, 80) in preprocessing , I think it will be easier to make test with no need for patching. Besides, I think this patching and average method used in testing may act like "ensemble learning", which will improve performance.

1. I think: first, this can increase the diversity of train data, and second, it can increase the robustness of the model (because even if we cut the preprocessing well, the data of the test set may not match completely in this range(112, 112, 80)).
2. "ensemble learning", I think you have answered your question.:smile:

[wtkszzz]

这个项目是一个很棒的框架。但我有一些问题要问。

1. 我注意到在预处理阶段，您根据标签掩码将每个图像裁剪成一个框（大于补丁大小（112、112、80））并获得一个 LA 区域。但后来，在 Dataset LAHeart 中，您添加了一个随机裁剪变换以再次将图像裁剪为 (112, 112, 80)。为什么这个？我认为只需裁剪到 (112, 112, 80) 并删除变换中的随机裁剪就足够了。随机裁剪可能会丢失一些 LA 信息。
2. 在测试阶段，为什么要对一张图像的补丁进行多次推断？如果在预处理中裁剪到 (112, 112, 80) ，我认为不需要打补丁就可以更容易地进行测试。此外，我认为在测试中使用的这种修补和平均方法可能会起到“集成学习”的作用，这将提高性能。

1. 我认为：第一，这样可以增加训练数据的多样性，第二，可以增加模型的鲁棒性（因为即使我们把预处理切得很好，测试集的数据也可能在这个范围内不完全匹配（112 , 112, 80))。
2. “集成学习”，我想你已经回答了你的问题。😄

你好！请问你用其它数据集跑过这个项目吗？