Unlike annotation format of segmentation in Youtube-VIS, which is RLE counts in number, this counts in segmentation annotation is string. Is this encrypted or I got wrong data downloaded?
To load the segmentation annotations, please refer to the annToMask function in https://github.com/youtubevos/cocoapi/blob/master/PythonAPI/pycocotools/mask.py .
The segmentation annotations in OVIS are in the compressed RLE format, which have been compressed by the maskUtils.frPyObjects function. These segmentation annotations can be directly loaded by the annToMask function.