Lacuna Dataset

[sourav-roni]

Hi, how should we go forward in replicating the results for Lacuna data-sets mentioned in the paper. Specifically, how to get the data in the place for the experiments?

Thanks in advance

[YuyangXueEd]

I think in the paper they mentioned that Lacuna-10 consists of face images of 10 celebrities from VG- GFaces2, randomly sampled with at least 500 images each. We split the data into a test set of 100 samples for each class, while the remaining form the training set. Similarly, Lacuna-100 randomly samples 100 celebrities with at least 500 images each. We resize the images to 32x32. There is no overlap between the two Lacuna datasets.