Data Normalization for BigBiGAN

[xiao7199]

Hi, Please check if the normalization is correct for the BigBiGAN encoder. Based on the description of (https://github.com/ali-design/GenRep/blob/master/utils/utils_bigbigan.py#L115), the input data for BigBiGAN should be normalized between [-1,1]. But, at L144-145 of main_linear_bigbigan_encoder.py, you use ((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)) instead of ((0.5,0.5,0.5), (0.5,0.5,0.5)) which maps images to [-1,1]. I tried your default option for main_linear_bigbigan_encoder.py on ImageNet100 , and it gives 54 Top1 ACC, close to 55.7, as reported in Sec 4.2.
However, when I use ((0.5,0.5,0.5), (0.5,0.5,0.5)), it gives me over 72 Top1 ACC.