Self-training method relies on making inference on unlabeled data and using created pseudo-labels in further training.
In comparison to well-known two- or one-stage detectors, DETR does not need to set the number of anchor boxes or even threshold for NMS algorithm.
Neural networks can be useful in case of detecting trash in image, but they require collecting a large and good quality data.
Differentiating waste instances under a single class label is challenging task
Evaluation metrics that data scientists should know