`SemanticSegmentationData` - zero-size array to reduction operation maximum which has no identity

[marisancans]

🐛 Bug

Running example from docs causes an exception I cannot understand.

To Reproduce

Paste the code from docs:

import torch

import flash
from flash.core.data.utils import download_data
from flash.image import SemanticSegmentation, SemanticSegmentationData

# 1. Create the DataModule
# The data was generated with the  CARLA self-driving simulator as part of the Kaggle Lyft Udacity Challenge.
# More info here: https://www.kaggle.com/kumaresanmanickavelu/lyft-udacity-challenge
download_data(
    "https://github.com/ongchinkiat/LyftPerceptionChallenge/releases/download/v0.1/carla-capture-20180513A.zip",
    "./data",
)

datamodule = SemanticSegmentationData.from_folders(
    train_folder="data/CameraRGB",
    train_target_folder="data/CameraSeg",
    val_split=0.1,
    transform_kwargs=dict(image_size=(256, 256)),
    num_classes=1,
    batch_size=4,
)

# 2. Build the task
model = SemanticSegmentation(
    backbone="mobilenetv3_large_100",
    head="fpn",
    num_classes=datamodule.num_classes,
)

# 3. Create the trainer and finetune the model
trainer = flash.Trainer(max_epochs=3, gpus=torch.cuda.device_count())
trainer.finetune(model, datamodule=datamodule, strategy="freeze")

# 4. Segment a few images!
datamodule = SemanticSegmentationData.from_files(
    predict_files=[
        "data/CameraRGB/F61-1.png",
        "data/CameraRGB/F62-1.png",
        "data/CameraRGB/F63-1.png",
    ],
    batch_size=3,
)
predictions = trainer.predict(model, datamodule=datamodule)
print(predictions)

# 5. Save the model!
trainer.save_checkpoint("semantic_segmentation_model.pt")

In the dataset folder (data) delete all files but leave first 4, set batch_size 1

Code sample

Expected behavior

Training with low number of samples. I understand that the problem could be that 4 images is not enough, but in my opinion the error message has to be more clearer. I currently have no idea what it means.

Environment

• OS (e.g., Linux): Ubuntu 20.04
• Python version: 3.9.15
• PyTorch/Lightning/Flash Version (e.g., 1.10/1.5/0.7): lightning-flash==0.8.1.post0 (latest pip install flash)
• GPU models and configuration: A2000
• Any other relevant information:

Additional context

Error message:

Exception has occurred: ValueError
zero-size array to reduction operation maximum which has no identity
  File "/mnt/data/src/word_bot/asd.py", line 12, in <module>
    datamodule = SemanticSegmentationData.from_folders(
ValueError: zero-size array to reduction operation maximum which has no identity

[Borda]

@marisancans could you pls check if the issue still exists on master? :chipmunk: