Spanish Sign Language (LSE) Fingerspelling Dataset

The dataset consists of over 160,000 images for Spanish Sign Language (Lengua de Signos Española or LSE) alphabet classification, also known as fingerspelling. The compressed files "192" and "512" correspond to the two image resolutions considered in the paper: 192x192 and 512x512, respectively. In each resolution folder, there are 3 subfolders that correspons to the three image variants considered: (1) Original, (2) Keypoints, and (3) Original-Keypoints, which correspond to the raw images, the keypoints with black background, and the the original images with added landmarks, respectively. The "Train" and "Validation" sub-folders contain the images that were used to train the models, separated by letter. There is also an "Evaluation" folder that contains the split that was used for evaluating the final models. Under CC BY 4.0 license, researchers can re-use this dataset, as long as you give appropriate credit by citing the dataset.