We complete all our annotation on an annotation platform.
-
Extract Filename and Choice:
- Parse each annotation result.
- Extract the
filename
from thefilePath
. - Extract the
choice
from theproperties
field in theevents
list.
-
Create Dictionary:
- Construct a dictionary where each key is the
filename
and the corresponding value is thechoice
.
- Construct a dictionary where each key is the
-
Prepare Motion Files:
- Follow the instructions in
motion.md
to process and prepare the corresponding motion.npz
files.
- Follow the instructions in
-
Pack into Datasets:
- Use the guidelines in
dataset.md
to pack the prepared motion files into datasets.
- Use the guidelines in
A typical annotation result for a video is structured as follows:
{
"filePath": "path_to_the_file/filename-ABCD.mp4",
"info": {
"duration": 3,
"framecount": 60,
"framerate": 20,
"width": 3840,
"height": 960
},
"events": [
{
"id": 1,
"event": "valid",
"startTime": 0.643,
"endTime": 0.98,
"properties": {
"Whether natural or not": "D"
}
}
]
}
The results indicate whether the annotation is valid and specify the choice from A/B/C/D. Then, the annotation results need to be reorganized into a dictionary with the format:
{
"filename": "choice"
}
Where choice
can be one of the following values: "A"
, "B"
, "C"
, "D"
, "allgood"
, or "allbad"
.
Then, you may refer to the motion.md
file for detailed information on how to handle and process motion files. Refer to the dataset.md
file for guidelines on how to prepare and pack the motion .npz
files into datasets.