This dataset is a modification of the D2S set to the COCO format.
The Annotations for the test set are not public and the results can be requested per mail if in the correct format.
The Augmentation set (10k images) is split into Train and Val atm (70-30) but we should maybe change this as the val set is important since we don't have the annotations of the test set
