- entityseg_train: Training data for the original Entity Segmentation task
- entityseg_train_01/02/03: same as above but split into 3 parts for ease of file loading
- entityseg_val: Validation data for the original Entity Segmentation task
- entityseg_train_lr: Low-resolution (800x1333) training data for the original Entity Segmentation task
- entityseg_val: Low-resolution (800x1333) validation data for the original Entity Segmentation task
- entityseg_insseg_train: Training data for the instance segmentation task built upon EntitySeg
- entityseg_insseg_val: Validation data for the instance segmentation task built upon EntitySeg
- entityseg_panseg_train: Training data for the panoptic segmentation task built upon EntitySeg
- entityseg_panseg_val: Validation data for the panoptic segmentation task built upon EntitySeg
- entityseg_semseg_train: Training data for the semantic segmentation task built upon EntitySeg
- entityseg_semseg_val: Validation data for the semantic segmentation task built upon EntitySeg
- url.json: JSON file that provides the URLs of all images considered by this dataset
For the annotation formats, we strictly follow the dataloader format of detectron2 in semantic, instance and panoptic segmentation. The format of entity segmentation follows the COCO instance segmentation format.
This dataset is released under Adobe Research License.