FastViT-autodistill error

[andysingal]

Search before asking

• I have searched the Autodistill issues and found no similar bug report.

Bug

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
[<ipython-input-4-c319f567610d>](https://localhost:8080/#) in <cell line: 17>()
     15 # base_model = FastViT(None)
     16 
---> 17 predictions = base_model.predict("example.png")
     18 
     19 labels = [FASTVIT_IMAGENET_1K_CLASSES[i] for i in predictions.class_id.tolist()]

6 frames
[/usr/local/lib/python3.10/dist-packages/torchvision/transforms/_functional_tensor.py](https://localhost:8080/#) in normalize(tensor, mean, std, inplace)
    926     if std.ndim == 1:
    927         std = std.view(-1, 1, 1)
--> 928     return tensor.sub_(mean).div_(std)
    929 
    930 

RuntimeError: The size of tensor a (4) must match the size of tensor b (3) at non-singleton dimension 0

Environment

pip install autodistill-fastvit roboflow transformers -Uq

Minimal Reproducible Example

from autodistill_fastvit import FastViT, FASTVIT_IMAGENET_1K_CLASSES
from autodistill.detection import CaptionOntology

# zero shot with prompts from FASTVIT_IMAGENET_1K_CLASSES
base_model = FastViT(
    ontology=CaptionOntology(
        {
            "Beagle": "beagle",
            "Border Collie": "collie"
        }
    )
)

# zero shot without prompts
# base_model = FastViT(None)

predictions = base_model.predict("example.png")

labels = [FASTVIT_IMAGENET_1K_CLASSES[i] for i in predictions.class_id.tolist()]

print(labels)

ERROR

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
[<ipython-input-4-c319f567610d>](https://localhost:8080/#) in <cell line: 17>()
     15 # base_model = FastViT(None)
     16 
---> 17 predictions = base_model.predict("example.png")
     18 
     19 labels = [FASTVIT_IMAGENET_1K_CLASSES[i] for i in predictions.class_id.tolist()]

6 frames
[/usr/local/lib/python3.10/dist-packages/torchvision/transforms/_functional_tensor.py](https://localhost:8080/#) in normalize(tensor, mean, std, inplace)
    926     if std.ndim == 1:
    927         std = std.view(-1, 1, 1)
--> 928     return tensor.sub_(mean).div_(std)
    929 
    930 

RuntimeError: The size of tensor a (4) must match the size of tensor b (3) at non-singleton dimension 0

Additional

No response

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

[capjamesg]

This error is now fixed. The issue was that the image had four channels -- RGBA -- and the module only supports RGB images. The requisite conversion code has been added to autodistill-fastvit, and we will likely incorporate similar logic in the main autodistill load_image function.

To use the updated code, run pip install --upgrade autodistill-fastvit.