The process_batch tool loads a YOLOv5 TensorRT engine, and performs object detection on a batch of images at once, i.e. batch inference. Note that the YOLOv5 model should have been exported with an explicit batch size.
Basic usage:
./process_batch --engine ENGINE_FILE --inputs INPUT_DIRECTORY --outputs OUTPUT_DIRECTORY
Arguments:
--engine: path to the YOLOv5 TensorRT engine--inputs: path to the directory in which the images are stored--outputs: path to the directory in which the results should be written (directory should exist already)--classes: (optional) path to a file containing the class names
By default, the process_batch program will attach numbers representing the class to all of the detections. For instance, when using the COCO dataset, this ranges from 0 to 79. By specifying the class names corresponding to each class id, human-readable names (e.g. "car", "truck") are displayed instead.
For convenience, a file containing the class names for COCO dataset can be found here.
Given that your YOLOv5 TensorRT engine is yolov5s.engine and your input images are stored in directory input_images, you can detect objects using:
./process_batch --engine yolov5s.engine --inputs input_images --outputs output_images
Visualizations will be stored to disk in the directory output_images.