Badminton shuttle tracking

Attempts to track the position of the shuttle(s) in a video.

Usage summary

# Dataset specific commands
# Run once to load data into appropriate locations

# Clean images and annotation files
python clean.py BADMINTON_DATA_ROOT

# Copy data to the preset location
python prepare.py --src BADMINTON_DATA_ROOT
# Alternatively, just copy a subset
python prepare.py --src BADMINTON_DATA_ROOT --include 01,02,05



# Training

# Test train split
python split.py

# Actual training and evaluation
python main.py
# Or start from an intermediate step with different arguments
python main.py --start_from eval --iou_thres 0.25

Data preparation

File layout

The paths for the input are specified by --frames_path and --xml_path. The files should have the following structure:

ROOT/
    VID1/
        frame_001
        frame_002
        ...
    VID2/
        frame_001
        frame_002
        ...
    ...

The filename doesn't matter as long as they are ordered correctly.

Test train split

For training, the data are split into training and testing sets. This is done by specifying the filenames in files --training_set and --testing_set.

The files can be created manually or generated with split.py.

Exclusion area

To avoid mis-labeling shuttles rolling on the floor as "non-ball", the program accepts coordinates of the area where shuttles might land on. The coordinates are specified in JSON format, e.g.

{
    "VID1": [x1, y1, x2, y2],
    "VID2": [x1, y1, x2, y2],
    ...
}

The default location of the JSON file is data/exclude.json