Here are some example scripts that make use of this repository.
Before running the scripts, install deeptangle
locally from the root directory
pip install -e .
as well as the requirements.txt
file using pip.
pip install -r requirements.txt
You also need to download the weights and the input videos.
The main purpose of the model is to detect splines. Thus, detect.py
provides a simple script to show how to do so.
An example usage would be:
python3 detect.py --model=weights/ --input=/path/to/video.avi
The flags are:
--correction_factor: Value of the correction_factor.
(default: '1.0')
(a number)
--frame: Target frame to detect
(default: '5')
(an integer)
--input: Path to the video.
--model: Path to the weights
(default: 'ckpt')
--output: File where the output is saved.
(default: 'out.png')
--overlap_threshold: Overlap score threshold to suppress predictions.
(default: '0.5')
(a number)
--score_threshold: Score threshold to prune bad predictions.
(default: '0.5')
(a number)
Likewise, a track.py
example is included, where the batching trick to increase performance is shown.
An example usage would be:
python3 track.py --model=weights/ --input=/path/to/video.avi
The flags are
--correction_factor: Value of the correction_factor.
(default: '1.0')
(a number)
--initial_frame: First target frame to start tracking.
(default: '5')
(an integer)
--input: Path to the video.
--model: Path to the weights
--num_batches: Maximum number of batches to do simultaneously.
(default: '10')
(an integer)
--num_frames: Number of frames to perform tracking.
(default: '0')
(an integer)
--overlap_threshold: Overlap score threshold to suppress predictions.
(default: '0.5')
(a number)
--score_threshold: Score threshold to prune bad predictions.
(default: '0.5')
(a number)
--output: Location to store the frames.
(default: 'out/')