Convert Gray Image to RGB format for EfficientAD not working

[Dpat123]

i am trying to convert GrayScale image to RGB for EfficientAD model learning but it is not working
i have tried following things but none of them working . during model learning it gives error . could you please suggest proper conversion for GrayScale images for use with Anomalib

from anomalib.data.utils.image import read_image, save_image
import numpy as np

            img = read_image(img_path)
            # If the image is 2D, convert it to 3D by stacking the channels
            if img.ndim == 2:
                img = np.stack((img,) * 3, axis=-1)

            # Unconditionally convert to float32 to ensure compatibility
            img = img.astype(np.float32)
            
            save_image(output_path, img)

[alfieroddan]

The read_image function already converts GrayScale images on disk to RGB:

 image = Image.open(path).convert("RGB")

anomalib/src/anomalib/data/utils/image.py

Line 300 in 3e2906a

def read_image(path: str | Path, as_tensor: bool = False) -> torch.Tensor | np.ndarray:

Can you show me your full code for running a model on your custom data? Is there an error?

Something like this:

from anomalib.data import Folder
from anomalib.engine import Engine
from anomalib.models import Padim

model = Padim()

datamodule = Folder( #<------- replace with your data here
    name="custom_folder",
    root="./datasets/custom",
    normal_dir="good",
    abnormal_dir="defect"
)

engine = Engine()

engine.fit(
    model=model,
    datamodule=datamodule,
)

[alfieroddan]

from anomalib.data import Folder
from anomalib.engine import Engine
from anomalib.models import Padim

model = Padim()

datamodule = Folder( #<------- replace with your data here
    name="custom_folder",
    root="./datasets/custom",
    normal_dir="good",
    abnormal_dir="defect"
)

engine = Engine()

engine.fit(
    model=model,
    datamodule=datamodule,
)

[Dpat123]

Folder only accept folder path it does not accept any sequence . so how i am going to use read_image and pass that to Folder ?

[alfieroddan]

From the DataModule documentation: https://anomalib.readthedocs.io/en/latest/markdown/guides/reference/data/datamodules/index.html

You can load your custom data in Folder or Tabular. Let me show you two ways quickly but next time have a scroll through the docs 😄.

Folder
https://anomalib.readthedocs.io/en/latest/markdown/guides/reference/data/datamodules/image/folder.html

The Folder data module expects your data to look like this:

root/
├── normal_dir/
│   ├── image1.png
│   └── image2.png
├── abnormal_dir/
│   ├── image3.png
│   └── image4.png
└── mask_dir/
    ├── mask3.png
    └── mask4.png

And then you can load like this:

datamodule = Folder( #<------- replace with your data here
    name="custom_folder",
    root="./datasets/custom",
    normal_dir="good",
    abnormal_dir="defect"
)

This is just a standard that Anomalib and detests like MvTec often use.

Tabular
https://anomalib.readthedocs.io/en/latest/markdown/guides/reference/data/datamodules/image/tabular.html

This is where you can use an image sequence. The samples dict contains MATCHING paths, label_indexes and split. So the first index of all three lists represent one image's path, label and split.

from anomalib.data import Tabular
from anomalib.data.utils import Split, LabelName

samples = {
    "image_path": ["images/image1.png", "images/image2.png", "images/image3.png", ... ],
    "label_index": [LabelName.NORMAL, LabelName.NORMAL, LabelName.ABNORMAL,  ... ],
    "split": [Split.TRAIN, Split.TRAIN, Split.TEST, ... ],
}
datamodule = Tabular(
    name="custom",
    samples=samples,
    root="./datasets/custom",
)

Hope that helps.

[Dpat123]

Thanks for the same i will try that.

[Dpat123]

I got this working for Gray Images . To setup DataSet as suggest above
`
dataset_root = "Anomalib/datasets/top_bottles/train"

normal_image_paths = sorted([
str(file_path.relative_to(dataset_root)) for file_path in Path(dataset_root +"/"+ "normal").iterdir()
])
anomalous_image_paths = sorted([
str(file_path.relative_to(dataset_root)) for file_path in Path(dataset_root +"/"+ "anomaly").iterdir()
])

samples = pd.DataFrame({
"image_path": normal_image_paths + anomalous_image_paths,    
"label_index": [0] * len(normal_image_paths) + [1] * len(anomalous_image_paths),
"split": ["train"] * len(normal_image_paths) + ["test"] * len(anomalous_image_paths),
})

Create a Tabular datamodule

datamodule = Tabular(
name="TopBottle",
samples=samples,
train_batch_size=1,
eval_batch_size=1,
root=dataset_root,
num_workers=1,
val_split_mode="from_test",  # Create validation set from test set
val_split_ratio=0.5,  # Use 10% of test set for validation
)
datamodule.setup()

Train images

train_sample = next(iter(datamodule.train_data))
val_sample = next(iter(datamodule.val_data))
print(train_sample.image.shape)
print(f"Number of validation samples: {len(datamodule.val_data)}")
print(val_sample.image.shape)

`