Figure 11.

Examples of successful instance detection from both the test:seen set (a) and the test:unseen set (b). The top images depict the first four channels of the neural network output. The middle image composed of six sub-images depicts the color-coded vector associations from the last 12 channels of the neural network output. The bottom images depict both ground truth locations and estimates using the following color coding: false negative (blue), and true positive (green). Note that these images depict 100% successful detections, so only true positives are present.