yanren1/ReinAD

--- license: cc-by-nc-sa-4.0 language: - en size_categories: - 100B<n<1T --- # ReinAD: Towards Real-world Industrial Anomaly Detection with a Comprehensive Contrastive Dataset Our dataset consists of a training set and a test set. All normal and anomaly images are in hdf5 format. In the mask annotations, pixels with a value of 0 represent normal regions, and pixels with a value of 1 represent anomaly regions. The file structure of the training set and the test set are consistent, as follows: ```text dataset/ ├── train/ │ ├── category1.h5 │ ├── category2.h5 │ └── ... │ └── test/ ├── category1.h5 ├── category2.h5 └── ... ``` The structure of the hdf5 file is as follows, where ```chunk_size = 100```: ```text / (root) ├── attrs │ ├── split: "train"/"test" │ └── category: category_name │ ├── Images │ ├── Anomaly_0: [chunk_size, H, W, C] # Anomaly images │ ├── Anomaly_1: [chunk_size, H, W, C] │ ├── ... │ ├── Normal_0: [chunk_size, H, W, C] # Normal images │ ├── Normal_1: [chunk_size, H, W, C] │ └── ... │ └── Masks ├── Anomaly_0: [chunk_size, H, W] # Pixel-level annotations for anomaly images ├── Anomaly_1: [chunk_size, H, W] └── ... ```