Input Modules - src.runtime.modules.input namespace

An input module is responsible to supply frames to the application. To achieve this the module gets the process_frames method passed, which is used by the module to send data to the FrameProcessor. See its documentation for more details on its parameters

The input module will only be called once. As soon as it returns the application will exit. An input module has to be registered in setup_input.

Submodules

src.runtime.modules.input.input_images module

src.runtime.modules.input.input_images.input_images(process_frames: Callable[[torch.Tensor, List[str], List[numpy.ndarray]], None], input_file='cfg.data_root/cfg.test_txt', data_root='cfg.data_root')

load images frame by frame and passes them to process_frame

used non-basic-cfg values:

• test_txt

Parameters

• process_frames – function taking a list of preprocessed frames, file paths and source frames

• input_file – index txt file to process

• data_root – root directory of dataset

src.runtime.modules.input.input_screencap module

src.runtime.modules.input.input_screencap.input_screencap(process_frames: Callable, mon: dict) → None

record from screen batch size is always 1

This is was implemented to test GTA. Its a bit difficult to use. You have to manually specify the position and size of your target window here. If your information are wrong (out of screen) you’ll get a cryptic exception! Make sure your config resolution matches your settings here.

used non-basic cfg options: screencap_enable_image_forwarding

Parameters

• process_frames – function taking a list of preprocessed frames, file paths and source frames

• mon – position and size of recording window, eg {‘top’: 0, ‘left’: 3440, ‘width’: 1920, ‘height’: 1080}

src.runtime.modules.input.input_video module

src.runtime.modules.input.input_video.input_camera(process_frames: Callable[[torch.Tensor, List[str], List[numpy.ndarray]], None], camera_number: int = 'cfg.camera_input_cam_number')

camera input wrapper for input_video()

used non-basic-cfg values:

• camera_input_cam_number

Parameters

• process_frames – function taking a list of preprocessed frames, file paths and source frames

• camera_number – opencv camera index

src.runtime.modules.input.input_video.input_video(process_frames: Callable[[torch.Tensor, List[str], List[numpy.ndarray]], None], input_file: Union[str, int] = 'cfg.video_input_file', names_file: str = None)

read a video file or camera stream. batch size is always 1

used non-basic-cfg values:

• video_input_file

Parameters

• process_frames – function taking a list of preprocessed frames, file paths and source frames

• input_file – video file (path; string) or camera index (integer)

• names_file – list with file paths to the frames of the video; if names_file and frames (jpg’s) are available the input images module can also be used