fcakyon / augmented-maskrcnn Goto Github PK

🐛 Bug

During the training, in some images, following error occurs:

Exception has occurred: RuntimeError
cannot reshape tensor of 0 elements into shape [0, -1, 4] because the unspecified dimension size -1 can be any value and is ambiguous
  File "/home/augmented-maskrcnn/core/engine.py", line 30, in train_one_epoch
    loss_dict = model(images, targets)
  File "/home/augmented-maskrcnn/train.py", line 188, in train
    train_one_epoch(model, optimizer, data_loader, device, epoch, print_freq=10)
  File "/home/augmented-maskrcnn/train.py", line 208, in <module>
    train()

To Reproduce

Steps to reproduce the behavior:

1. set configuration
2. start training
3. get exception during evaluation

Expected behavior

Should continue training without any issues

Environment

• Python version (e.g., 3.6/3.7): Python 3.6
• OS (e.g., Linux/Windows/MacOS): Ubuntu 18.04

I'm training to train with roof data but the training never starts... probably is in the augmenting stage... but it's take forever...

What can I do to debug that?

Hi !
First, great implementation, much better than most out there when it comes to code clarity.

I am running into some small issues:
When dataset entries are handed of too albumentations, albumentations throws a value array because the bbox of an annotation
is too small. It seems to me that after checking my dataset that this comes from numerical imprecision when rastering the mask polygons and calculating the bbox based on those ... Any advice ?

Secondly, the code assumes that category_ids are continuous from 0 to x ... i can work around that by remapping them, but there are a lot of places in the codebase where this is used and so far i have not managed to find one central place where to do the remapping so that it gets picked up by all other parts of the codebase. Our category ids come from a licensed SAAS tool for image labeling, and as such we have to remap them after exporting from there, as we cannot change the SAAS providers code ..