Detect¶

Synopsis¶

python  detect.py   [-p paths] [--expected_duration] [-c config]
                    [-w weights] [--conf] [--iou] [--half] [--force]
                    [--overwrite]

Description¶

If you have converted your video files to one of the supported file formats (.avi,.mkv,.mov,.mp4) or you already have such video files, you are ready to start the detection of the road users in each video frame.

Parameters¶

paths (required)¶

Filename convention

To prevent to set the start date and time in the config file for each individual video file, the current version of OTVision reads the start date and time from the video filenames.

Video files recorded by OTCamera already contain the start date and time in the filename.

In case you use OTVision to process video files recorded by other camera systems, please make sure that the filename of these files contain the start date and time in the following format: YYYY-MM-DD_hh-mm-ss

-p "path/to/video files" "path/to/other video files"

or

--paths "path/to/video files" "path/to/other video files"

One or multiple paths to video files or folders containing video files.

This parameter is required to run detect.py. It has to be specified either using the CLI or in the configuration yaml file.

expected_duration (required)¶

--expected_duration <video duration [sec]>

Expected duration of each video in seconds (must be all the same). This parameter helps to avoid errors if some images are missing in a video.

This parameter is required to run detect.py. It has to be specified either using the CLI or in the configuration yaml file.

config¶

-c "path/to/config file"

or

--config "path/to/config file"

Path to a custom user configuration yaml file. Other parameters (including paths) are parsed from this configuration file.

This parameter is optional. By default, the user_config.otvision.yaml in the root directory is read. This also overrides the default values of all parameters. If the user_config.otvision.yaml in the root directory is missing and no other configuration file is specified, the default values of the parameters as described below are used.

Warning

Any parameter passed to the CLI will overwrite the respective parameter from the config file.

weights¶

-w <weights> or --weights <weights>

Name of weights from PyTorch hub or path to weights file.

This parameter is optional and defaults to yolov8s.pt.

conf¶

--conf <float>

The YOLOv8 model confidence threshold. Should be a float value between zero and one.

The confidence threshold is the minimum confidence for a detection to be considered a true detection (otherwise this detection will be ignored entirely).

This parameter is optional and defaults to 0.25.

iou¶

--iou <float>

The YOLOv8 model IOU threshold. Should be a float value between zero and one.

The IOU threshold is the overlap threshold for areas of bounding boxes used in non-maximum suppression to avoid duplicate detections.

This parameter is optional and defaults to 0.45.

half¶

--half to use half precision (FP16) to speed up detection.

--no-half to not use half precision.

This parameter is optional and defaults to --no-half.

Half precision only works for running detection on a GPU!

force¶

--force to force a reload of a YOLOv8 standard model from PyTorch hub instead of using a cached model from previous detection runs.

--no-force to prevent forcing this reload.

This parameter is optional and defaults to --no-force.

overwrite¶

--overwrite to overwrite existing .otdet files.

--no-overwrite to prevent overwriting existing .otdet files.

This parameter is optional and defaults to --overwrite.