>Object detection from peripheral vision is not difficult though. Because we understand high level semantic information, which is largely robust against blurri

>Object detection from peripheral vision is not difficult though. Because we understand high level semantic information, which is largely robust against blurring. >The random filtered images looked much more corrupted to me than the blurred ones, which is consistent with blurred images being part of the training for the human visual system, It's also consistent with the neural network training on surface level regularities, which a random filter would corrupt at a much lower rate than a blur filter would.