search

JingweiJ / ActionGenome

A video database bridging human actions and human-object relationships
MIT License
127 stars 17 forks source link

Question about the dataset #5

Open ZihaoZheng98 opened 4 years ago

ZihaoZheng98 commented 4 years ago

Hi,I have a small doubt:For a frame that is annotated,how to determine the objects?Did you first extract the list of objects included in all the actions of a video, and then label each frame by the annotator to determine which objects in the list are included in the frame?Or you just use the objects occured in the actions of which the interval contains the frame?

JingweiJ commented 4 years ago

Hi, we started with the action interval annotations from Charades dataset. Note that each action is associated with an object class (e.g. "watching tv" is associated with "tv"). Within each action interval, we sample 5 frames uniformly, and on the sampled frames, we ask the annotator to label where the associated objects are (bounding boxes) and what the relationships are between the objects and the person.

TAA66 commented 4 years ago

Hi, I also have some questions regarding the annotations. For each sampled frame, the information about the associated objects are included, but not the action itself. So I wonder how I can find out which action class each annotated frame corresponds to? Is there, for example, a mapping from actions to their associated objects? @JingweiJ