Segmentation training fails

[till-ko]

The Segmentation training fails with the following output:

$ python3 -m weed_annotator.full_pipeline.run_full_pipeline
2023-04-13 15:03:05,730 Generating image compositions for training.
2023-04-13 15:03:24,496 Training semantic segmentation model on: sample_data/synthetic_imgs/10_no_leave_augmentations.
Downloading: "https://download.pytorch.org/models/resnet18-5c106cde.pth" to /home/dennis/.cache/torch/hub/checkpoints/resnet18-5c106cde.pth
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 44.7M/44.7M [00:03<00:00, 13.8MB/s]
Traceback (most recent call last):
  File "~/anaconda3/envs/weedannotator/lib/python3.6/runpy.py", line 193, in _run_module_as_main
    "__main__", mod_spec)
  File "~/anaconda3/envs/weedannotator/lib/python3.6/runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "~/WeedAnnotator/weed_annotator/full_pipeline/run_full_pipeline.py", line 48, in <module>
    train(train_config)
  File "~/WeedAnnotator/weed_annotator/semantic_segmentation/train.py", line 73, in train
    train_loader = DataLoader(train_dataset, batch_size=config["training"]["batch_size"], shuffle=True, num_workers=1)
  File "~/anaconda3/envs/weedannotator/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 262, in __init__
    sampler = RandomSampler(dataset, generator=generator)  # type: ignore
  File "~/anaconda3/envs/weedannotator/lib/python3.6/site-packages/torch/utils/data/sampler.py", line 104, in __init__
    "value, but got num_samples={}".format(self.num_samples))
ValueError: num_samples should be a positive integer value, but got num_samples=0

We tried to run the sample (unchanged from this repository) in a docker container using python 3.6, a venv using python 3.6 as well as in a conda environment. Same Issue everytime. Possible Issue with the DataLoader?

It seems like the generated images are not forwarded to be reused in training. Are we supposed to make changes to the config?