Custom dataset

[LorenzoF6]

Hi, if i want to test some models with my own custom dataset, it should be placed in a specific directory in the anomalib library o can i create my own folder called Datasets (I'm on divre and work with Colab), put my custom datasets and give the absolute path of the position of the dataset?

And if i have different type of defect, it is correct to construct my custom dataset following a mvtec like structure ?

Thansk

[yxteo2]

For me I make my data in this format in a folder:

├── train
├── good
├── 000.png

├── test
├── Defect1
├── 000.png
├── Defect2
├── 000.png

├── ground_truth
├── Defect1
├── 000_mask.png
├── Defect2
├── 000_mask.png

With this I can use the engine.config to fit the model, where the data path is the data to this folder

[yxteo2]

Yaya, I think you can select the format by adjusting the data class_path in the yaml file, you can select other format also

data:
class_path: anomalib.data.MVTec

[LorenzoF6]

so the idea is modify the .yaml file and using cli command to train e do the rest.
But if i wanti to use the api such has the folder class (and its datamodule) how i mangaed to gave the correct path ?

[yxteo2]

If I not mistaken you can use the Folder function

#2311

[LorenzoF6]

so in order to get my own dataset path i have to modify the mvtec.yaml e folder.yaml and put in default root dataset the path to the dataset? This work only for the cli or also the api?
Thansk for all the help me to understand

[yxteo2]

so in order to get my own dataset path i have to modify the mvtec.yaml e folder.yaml and put in default root dataset the path to the dataset? This work only for the cli or also the api? Thansk for all the help me to understand

Yes, you need to modify the mvtec.yaml file
If you are using Folder dataset, you can adjust class_path to anomalib.data.folder