Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Bug with masks

See original GitHub issue

🐛 Bug

The masks do not follow the rotation of the images

To Reproduce

Steps to reproduce the behavior:

1.Define the following transformation:

from albumentations import (
    PadIfNeeded,
    HorizontalFlip,
    VerticalFlip,
    CenterCrop,    
    Crop,
    Compose,
    Transpose,
    RandomRotate90,
    ElasticTransform,
    GridDistortion, 
    OpticalDistortion,
    RandomSizedCrop,
    OneOf,
    CLAHE,
    RandomBrightnessContrast,    
    RandomGamma,
    RandomScale,
    LongestMaxSize
)
from albumentations.pytorch.transforms import ToTensor as AlbToTensor

inputs_transformations = Compose([
    RandomScale(scale_limit=0.3, p=0.5),
    LongestMaxSize(256),
    HorizontalFlip(p=0.3),
    VerticalFlip(p=0.3),
    CLAHE(p=0.5),
    RandomBrightnessContrast(p=0.5),    
    RandomGamma(p=0.5),
    AlbToTensor()])

I am using kaggle penumonothorax dataset The result of applying three different transformations to the image and the mask can have the following result:

As you can see, the image is horizontally flipped, whereas the mask is vertically filpped.

Version I use is from pip: 0.3.0

The same error can be seen when applying Flip transform

Issue Analytics

State:
Created 4 years ago
Comments:6

Top GitHub Comments

2reactions

albucommented, Jul 31, 2019

yeah, we will consider supporting non-rgb or even add warning to all transforms which does not support grayscale

0reactions

pyafcommented, Jul 31, 2019

This is a subtle bug, spent an hour figuring it out. There should be a warning for non three channel cases. As of now, it returns the transposed image (weird, yeah) @albu what do you think?