Object detection is used to locate and identify objects in images. Azure´s "custom vision service" was used to train a model to recognize specific classes of objects in images.
- set up a python 3.8 environment
- pip install azure-cognitiveservices-vision-customvision==3.1.0
Images of three types of objects were uploaded to the custom vision ai tool for tagging and subsequent training.
After initial model training, the model was further improved by uploading more tagged training images using the python sdk i.e. running script "train-detector.py". Despite a small set of tagged images, the ai performs already very well with a recall of 93% and a precision of 100%.
The image shown below was uploaded for inference to test the ai model by running the script "test-detector.py. As can be seen, the ai model detectes all three items with scores clearly above 90%. Also the objects locations seem very accurate.