We provide a single-file demo in this folder to facilitate getting started. Supposing that you are in the root directory of this project.

To use the demo, you need first generate a class name list to tell the OMG-Seg model all possible categories as the vocabulary dictionary. We have already provided a sample vocabulary list in demo/configs/names/th139_st101.py.

Then, we need to generate the class embeddings based on the names. You can do this by the following command:

PYTHONPATH=. python tools/gen_cls.py demo/configs/m2_convl.py

The script will automatically read the class list, which is imported in demo/configs/m2_convl.py (please refer to CLASSES and DATASET_NAME), and generate the embeddings.

After generating the embeddings, you can run the demo by:

PYTHONPATH=. python demo/image_demo.py

for image; and

PYTHONPATH=. python demo/video_demo.py

for video.

Please refer to test_image and test_video for the visualization of the outputs.

If you want to try your own images or videos, please change the IMG_PATH, VID_PATH, and MODEL_PATH.

If you want to customize our model, please refer to the config scripts (demo/configs/m2_convl.py and demo/configs/m2_convl_vid.py) for details.

Note that all the model-related code have been imported in the config file. You need to find the corresponding path to find the model implementation details.