Now that the project uses the MSCOCO class to load the dataset from the annotation json file, the way files are loaded from disk has become more or less obsolete and unnecessary. It should be converted to a dataset class, like MSCOCO currently is ('json' mode works fine), so that they can share functionality.
Now that the project uses the MSCOCO class to load the dataset from the annotation json file, the way files are loaded from disk has become more or less obsolete and unnecessary. It should be converted to a dataset class, like MSCOCO currently is ('json' mode works fine), so that they can share functionality.
Objections and/or suggestions are welcome.