C1M3 - Unittest failing on std deviation of datase

Course Q&A Professional Certificate:PyTorch for Deep Learning PyTorch: Fundamentals
,
1

Hi,

I get the unit test 3 failing with:
Failed test case: The training dataset does not have the correct transform.
Expected: The training dataset should use the `augmentation_transform`: Compose(
RandomVerticalFlip(p=0.5)
RandomRotation(degrees=[-15.0, 15.0], interpolation=nearest, expand=False, fill=0)
Resize(size=(128, 128), interpolation=bilinear, max_size=None, antialias=True)
ToTensor()
Normalize(mean=[0.6659, 0.6203, 0.4784], std=[0.2119, 0.2155, 0.2567])
).
Check that the training dataset is wrapped with `SubsetWithTransform` using `augmentation_transform`.
Got: The training dataset transform is Compose(
RandomVerticalFlip(p=0.5)
RandomRotation(degrees=[-15.0, 15.0], interpolation=nearest, expand=False, fill=0)
Resize(size=(128, 128), interpolation=bilinear, max_size=None, antialias=True)
ToTensor()
Normalize(mean=tensor([0.6659, 0.6203, 0.4784]), std=tensor([0.2888, 0.2884, 0.3426]))
).

The problem seem due to different std but std is calculated with prebuilt code that I didn’t change. It affects my final score (even if I managed to complete the module as well)

2

Hi @Giustino,

I’m going to send a direct message regarding this.

Also, this is not C1M4, this is from C1 M3 assignment. I have edited your topic title to reflect that.

3

Hi @Giustino,

Reading through the error, it mentions:

Check that the training dataset is wrapped with `SubsetWithTransform` using `augmentation_transform`.

As per the exercise instructions, you are expected to use augmentation_transform (the function parameter).

Please make sure you are using that, not any other variable, and your test should pass.

Best,
Mubsi

1 Like
4

Thank you very much, it solved the problem