A Novel Attentive Generative Adversarial Network for Waterdrop Detection and Removal of Rubber Conveyor Belt Image

Figure 7

Results of comparing a few different methods on the rubber conveyor belt image of the water droplet shape of stilliform water droplets, and figures show in the sequence input, ground truth, DSC results, LP results, CNN results, ATTGAN results, our results detected waterdrop attention map. Nearly all water droplets are removed by our method despite the diversity of their colors, shapes, and transparency. (a) Input 1. (b) Ground truth. (c) DSC. (d) LP. (e) CNN. (f) ATTGAN. (g) Ours. (h) Attention map. (i) Input 2. (j) Ground truth. (k) DSC. (l) LP. (m) CNN. (n) ATTGAN. (o) Ours. (p) Attention map.