search

uci-cbcl / NoduleNet

[MICCAI' 19] NoduleNet: Decoupled False Positive Reduction for Pulmonary Nodule Detection and Segmentation
Other
185 stars 57 forks source link

Training Data Problems #10

Closed Amadeuszhao closed 4 years ago

Amadeuszhao commented 4 years ago

Recently, when I tried to reproduce the result of your work(https://github.com/uci-cbcl/NoduleNet) , I encounted some issues. I modified your code and then followed your instruction step by step. And here’s the problem image As you can see there are 3498 files under the preprocess_data folder after I followed your preprocess instruction However when I strared training, image

There are only 62 training loaders, is this normal? At your convenience, would you please give me some of your insights. Thank you, I will appreciate it a lot.

tanghaotommy commented 4 years ago

Yes, it is normal. Each training loader iteration uses batch of images as input. The default batch size is set to 16, so each train iteration uses 16*62=992 nodules

Amadeuszhao commented 4 years ago

Yes, it is normal. Each training loader iteration uses batch of images as input. The default batch size is set to 16, so each train iteration uses 16*62=992 nodules

But I suppose the whole LUNA datasets have mode nodules.