miquelmarti/okutama-action

Okutama-Action: An Aerial View Video Dataset for Concurrent Human Action Detection

okutama-action's Issues

Google Drive links not working

Google made an update to some Google Drive file links, and I think the links in this repo were affected. Could you update those links (training set, test set, trained models)?

From what I can tell, it's necessary to recreate the link from Google Drive. The ID will be the same, but the link will have an extra parameter ?resourcekey=xxxxxxxxxxxx.

Data and label correspondence issues

Does anybody use this datesets? I am confused about the relationship between videos and labels. If i wanna use pictures to train my models, how to use the labels. Or someone have already used VATIC, but i was stucked in installing VATIC. Any advice would be greatly appreciated!

poor annotation quality

There are some annotations, taking bounding boxes as an instance, that are pretty precise in some videos like 1.1.2, 1.2.1, 2.1.2, 2.2.2, et al. While there still remains a large number of poorly annotated bounding boxes in other videos. Taking 2.1.1 for an example, during this video the 20th frame is well annotated, but the 140th, 350th frame and so many other frames are ridiculously wrong. Is there anything wrong with the annotation files that I have downloaded from http://okutama-action.org/ ? Could you please tell me how to correct these wrong annotations?

20th frame

140th frame

350th frame

action detection program mentioned in the paper

Can you please release the program for action detection that is mentioned in your paper?

Recommended way of bulk download

First, thank you the authors for curating the data set.

The data is currently on Google Drive under different folders. What is the recommended way to do a bulk download of the whole data set, without manually downloading each individual file? Google's zipping on-demand seems to take forever.

Regarding the camera meta file

Do you have any plan for releasing the camera meta file (i.e. altitude value)

Any concrete implementation ?

Hi, can you please provide some detailed implementation.