Skip to content

Commit

Permalink
Added notes for prepared data
Browse files Browse the repository at this point in the history
  • Loading branch information
JunweiLiang committed Jun 23, 2019
1 parent 20d4af5 commit b6b563c
Showing 1 changed file with 1 addition and 1 deletion.
2 changes: 1 addition & 1 deletion NOTES.md
@@ -1,7 +1,7 @@
## Prepared Data
Here are some notes on the prepared data's format, which could be useful for data visualization. The folder `next-data/final_annos/actev_annos` includes the following folders:

- `virat_2.5fps_resized_allfeature/`: This folder includes each data split's trajectories. Each video has one text file. The format is similar to the data in [Social-GAN](https://github.com/agrimgupta92/sgan). Each row is `frame_num person_id X Y`. The frame_num is 0-indexed. The person_id is unique within each video. You can see the frame_num is not continuous since we drop 12 frame every time to down-sample the frames so we have 2.5 FPS as the ETH/UCY data. The X, Y coordinates are **under 1920x1080 scale**, which means we resize the frames from scene `0002` in the ActEV data. The same goes to all other coordinates of bounding boxes and keypoints.
- `virat_2.5fps_resized_allfeature/`: This folder includes each data split's trajectories. Each video has one text file. The format is similar to the data in [Social-GAN](https://github.com/agrimgupta92/sgan). Each row is `frame_num person_id X Y`. The frame_num is 0-indexed. The person_id is unique within each video. You can see the frame_num is not continuous since we drop 12 frame every time to down-sample the frames so we have 2.5 FPS as the ETH/UCY data. The X, Y coordinates are **under 1920x1080 scale**, which means we have resized the 1280x720 frames from scene `0002` in the ActEV data. The same goes to all other coordinates of bounding boxes and keypoints.
- `anno_person_box/`: This folder includes each data split's person box coordinates. Each video has one pickle file. The pickle file is a dictionary with keys in the format of `${frame_num}_${person_id}` and the box coordinates are in the format of `x1 y1 x2 y2`.
- `anno_kp/`: Similar but with person keypoint coordinates in [17, 3] shape. The 17 keypoint indexes correspond to the ones in Detectron. Each row is `X Y logit`. Logit is in the range from 0 to 6 according to AlphaPose.
- `anno_activity/`: Similar but with each person's future and current activities: `current_actid_list, _, future_actid_list, _ = activity_pickle[key]`.
Expand Down

0 comments on commit b6b563c

Please sign in to comment.