MAuDe

Model Assertions for object Detection networks.

Running It

To classify things off the webcam, run:

cd /src
python live.py --model path/to/model/directory --confidence 0.4

Requirements

You need OpenCV, version 3.4 or greater. Easily installable via pip.

Models

You'll have to provide models yourself, as the weight files are pretty big. Take a look at YOLO for a model and how to install it.

Models are stored in a directory containing a configuration file config.toml. Each configuration file has 3 entries, all of which are paths to files: name references the name file, which is a label index, cfg, which hosts the network configuration, and weights, which has the network weights. An example file for YOLO v3 is given below:

names="coco.names"
cfg="yolov3.cfg"
weights="yolov3.weights"

This is a flat configuration, e.g. we expect coco.names to appear in the same directory, i.e.:

/models
+----/yolo
|    +----config.toml
|    +----coco.names
|    +----yolov3.cfg
|    +----yolov3.weights
+----/yolo-tiny
|    +----config.toml
...

Resources

We're based on OpenCV, although we use YOLO as our object detection model.