demo

Running this demo

1. Build the docker image and tag it. In this case we will call it saty/muscima.

   docker build MusicObjectDetector/demo/ -t saty/muscima

2. All hand writen music sheets should then be put in a single directory. In this example the directory will be /sheet_music.

3. Run the docker image with a volume mounted to the directory mapped to /workspace.

   # For invoking nvidia-docker
   nvidia-docker run --rm -t -v /sheet_music/:/workspace saty/muscima
   
   # For the new method of invoking the nvidia-docker runtime
   docker run --gpus all --rm -t -v /sheet_music/:/workspace saty/muscima

   Argument	Description
   run	Runs an image
   nvidia-docker or --gpus all	Use the nvidia runtime and allow access to GPUs
   --rm	Automatically removes the container once the command has finished
   -t	Allocates a pseudo-tty
   -v /sheet_music/:/workspace	Mounts /sheet_music/ to /workspace inside the container

4. Inference will now be performed on all image files in the /sheet_music directory. The results will be returned to /sheet_music/output

Using optional arguments

The following arguments are available:

Argument	Description
-h, --h	Shows a help message and exit
--detection_inference_graph DETECTION_INFERENCE_GRAPH	Specify a path to the frozen inference graph
--input_directory INPUT_DIRECTORY	Specify an input directory. In this use case, this must be the mapping used for the mounted volume
--detection_label_map DETECTION_LABEL_MAP	Specify a path to the detection label map
--output_dir OUTPUT_DIR	Specify a path for the output. If the path does not point to somewhere within the mounted volume, the results will be lost

To use these arguments, simply append the following docker commands:

docker run -rm -v /sheet_music/:/workspace saty/muscima [OPTIONS]

Note that there are no default arguments. The usage of any argument will thus require the use of all arguments

Here, we assume that you want to still perform inference on files in the /sheet_music directory. The /sheet_music directory is then mapped to /workspace/ in the docker container.

For further reference on running optional arguments, refer to the official documentation for docker run.

Downloading the trained model

A model pretrained on MUSCIMA++ can be downloaded from the releases page