Some of our main findings include that supplementing training on degraded images improve significantly the detection results. In low degradation scenarios, the improvement is better than the baseline ...