Curate your training data using a simple visual interface. Don't waste your time in labeling images which don't add value to training your model. Use the software to find the most relevant samples to label.
We started WhatToLabel after working on an image segmentation project where we had to annotate all the images. If you have to create pixel-perfect segmentation and are sitting in front of 10k frames you have a lot of work to do. We were not able to annotate all the frames (we spent up to 30 min per frame for annotation) so we needed a simple way to select "What To Label". Because we didn't find anything out there we built our own solution. WhatToLabel provides different state-of-the-art methods for data selection and works on unlabeled data. After finishing our project we saw that others struggled with the same problem too. There is simply way more data than we could ever annotate.
Lightly
Snipd