youtubevos / MaskTrackRCNN

MaskTrackRCNN for video instance segmentation based on mmdetection
Apache License 2.0
432 stars 75 forks source link

[Youtube-VIS] The category_id of the JSON file is not clear. #60

Open lihaoliu-cambridge opened 1 year ago

lihaoliu-cambridge commented 1 year ago

I am trying to build a JSON file for my own dataset in Youtube-VIS format. But in the "annotation" part, the key "category_id" is always the same for all instances. Could you please illustrate a little bit about this?

Since your segmentations and bboxes are lists for all frames in a video. Did you put the same instance across different frames in the same annotation, which shares the same "category_id"? If this is the case, for multiple instance data, how can I get the instance trajectory information across all frames for my own data?

Another explanation is that in the Youtube-VIS dataset, there is only one instance for all video. Hence, the key category_id is not working.

    {
        "info" : info,
        "videos" : [video],
        "annotations" : [annotation],
        "categories" : [category],
    }
    video{
        "id" : int,
        "width" : int,
        "height" : int,
        "length" : int,
        "file_names" : [file_name],
    }
    annotation{
        "id" : int, 
        "video_id" : int, 
        "category_id" : int, 
        "segmentations" : [RLE or [polygon] or None], 
        "areas" : [float or None], 
        "bboxes" : [[x,y,width,height] or None], 
        "iscrowd" : 0 or 1,
    }
    category{
        "id" : int, 
        "name" : str, 
        "supercategory" : str,
    }