bvr_web_pup

BVR Web Pup is an endpoint server that performs object detection on images on incoming images and returns back the results.

THIS PROJECT IS IN ALPHA. It should work, but don't expect too much yet.

Building

You need to have bvr_detect cloned and available for this project.

Assuming they are placed next to each other like this:

..
bvr_detect\
bvr_web_pup\

Your Cargo.toml should have:

bvr_detect = { version = "0.2.0", path = "../bvr_detect" }

Installation

Currently, there's no installation process. You would just run the binary with your parameters.

Building

cargo build --release

Running

./BvrWebPup <options>

Options you can set:

Description	Flag	Input
Path to weights/model file	-m --model-path	File path
Path to labels file	-c --classes-path	File path
Path to OnnxRuntime library	-l --lib-path-ort	Path to .so library file
Device to use for inference	-d --device	CPU, CUDA, TensorRT
Detection processing method to use	-r --processing-type	ORT, Torch, Python
Specify YOLO Version	-y --yolo-version	v4, v5, v6, v7, v8, v9, v10, v11
Split extra wide images (dual lens cameras) to process each half individually	-s	<bool flag>
Set detection threshold (default is 0.4)	-t --threshold	0.1 - 1.0 (higher is stricter)
Input width for the model	-w --width	A value divisible by 32
Input height for the model	-e --height	A value divisible by 32
Set a manual port for the web server (default 3000)	-p --port	Any available port on the system
Specify logging level (default INFO)	--log-level	ERROR, WARN, INFO, DEBUG, TRACE

Note: The case for the inputs is not important. Everything is converted to lowercase when evaluating.

TODO:

• Get this code published
• Ability to set thresholds for each camera, and each object per camera
• Web interface to see detection results and modify settings
  • Web interface
  • Option to save/view previous detections
    • Ability to easily re-process images with different settings for testing
  • Page to test detections with uploaded image