hendrycks/robustness

Maybe more detailed robustness results on AlexNet?

yuxinWen opened this issue · 3 comments

We're interested in studying the behavior of DNNs on ImageNet-C in more detail. However, we noticed that we cannot find the robust results of severity at each level for AlexNet even in Table 2. So could you provide these results at your convenience? Thanks!

Here are the AlexNet error rates at different severities.

Gaussian
=Average(0.69528, 0.82542, 0.93554, 0.98138, 0.99452)
Shot
=Average(0.71224, 0.85108, 0.93574, 0.98182, 0.99146)
Impulse
=Average(0.78374, 0.89808, 0.9487, 0.9872, 0.99548)

Defocus
=Average(0.656239999999999, 0.73202, 0.85036, 0.91364, 0.94714)
Glass
=Average(0.64308, 0.75054, 0.88806, 0.91622, 0.93344)
Motion
=Average(0.5843, 0.70048, 0.82108, 0.8975, 0.92638)
Zoom
=Average(0.70008, 0.769919999999999, 0.80784, 0.84198, 0.87198)

Snow
=Average(0.71726, 0.88392, 0.86468, 0.9187, 0.94952)
Frost
=Average(0.6139, 0.797339999999999, 0.8879, 0.89942, 0.9343)
Fog
=Average(0.67474, 0.7605, 0.84378, 0.8726, 0.945)
Brightness
=Average(0.4514, 0.48502, 0.54048, 0.62166, 0.724399999999999)

Contrast
=Average(0.64548, 0.7615, 0.88874, 0.9776, 0.9927)
Elastic
=Average(0.52596, 0.70116, 0.55686, 0.64076, 0.80554)
Pixelate
=Average(0.52218, 0.5462, 0.737279999999999, 0.87092, 0.91262)
JPEG
=Average(0.510019999999999, 0.54718, 0.57294, 0.654579999999999, 0.74778)

Speckle Noise
=Average(0.66192, 0.7444, 0.90246, 0.94548, 0.97268)
Gaussian Blur
=Average(0.54732, 0.70444, 0.82574, 0.89864, 0.9594)
Spatter
=Average(0.47196, 0.621939999999999, 0.75052, 0.84132, 0.90182)
Saturate
=Average(0.59342, 0.65514, 0.51174, 0.70834, 0.8226)

Thanks a lot!

Here are the AlexNet error rates at different severities.

Gaussian
=Average(0.69528, 0.82542, 0.93554, 0.98138, 0.99452)
Shot
=Average(0.71224, 0.85108, 0.93574, 0.98182, 0.99146)
Impulse
=Average(0.78374, 0.89808, 0.9487, 0.9872, 0.99548)

Defocus
=Average(0.656239999999999, 0.73202, 0.85036, 0.91364, 0.94714)
Glass
=Average(0.64308, 0.75054, 0.88806, 0.91622, 0.93344)
Motion
=Average(0.5843, 0.70048, 0.82108, 0.8975, 0.92638)
Zoom
=Average(0.70008, 0.769919999999999, 0.80784, 0.84198, 0.87198)

Snow
=Average(0.71726, 0.88392, 0.86468, 0.9187, 0.94952)
Frost
=Average(0.6139, 0.797339999999999, 0.8879, 0.89942, 0.9343)
Fog
=Average(0.67474, 0.7605, 0.84378, 0.8726, 0.945)
Brightness
=Average(0.4514, 0.48502, 0.54048, 0.62166, 0.724399999999999)

Contrast
=Average(0.64548, 0.7615, 0.88874, 0.9776, 0.9927)
Elastic
=Average(0.52596, 0.70116, 0.55686, 0.64076, 0.80554)
Pixelate
=Average(0.52218, 0.5462, 0.737279999999999, 0.87092, 0.91262)
JPEG
=Average(0.510019999999999, 0.54718, 0.57294, 0.654579999999999, 0.74778)

Speckle Noise
=Average(0.66192, 0.7444, 0.90246, 0.94548, 0.97268)
Gaussian Blur
=Average(0.54732, 0.70444, 0.82574, 0.89864, 0.9594)
Spatter
=Average(0.47196, 0.621939999999999, 0.75052, 0.84132, 0.90182)
Saturate
=Average(0.59342, 0.65514, 0.51174, 0.70834, 0.8226)

Could you please provide the result of AlexNet on ImageNet-P?
Thanks