annotation.md

Annotation Process and Dataset Preparation

Introduction

We complete all our annotation on an annotation platform.

Steps for Reorganization and Dataset Preparation

1. Extract Filename and Choice:

   • Parse each annotation result.
   • Extract the filename from the filePath.
   • Extract the choice from the properties field in the events list.

2. Create Dictionary:

   • Construct a dictionary where each key is the filename and the corresponding value is the choice.

3. Prepare Motion Files:

   • Follow the instructions in motion.md to process and prepare the corresponding motion .npz files.

4. Pack into Datasets:

   • Use the guidelines in dataset.md to pack the prepared motion files into datasets.

Usage

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.