The related code is found on line 47 of efficientnet_x_builder.py. According to equation (3) of the paper, the image size of efficientnet-b7 is 600, so

r=γ^Φ=1.15^Φ=600/224=2.6786

Φ=log(2.6786)/log(1.15)=7.04

then a conclusion can be drawn as

depth_coefficient=d=α^Φ=1.2^7.04=3.6094≠3.1

I have browsed several repos about EfficientNet, which most of them define depth of efficientnet-b7 as 3.1, so I'm not sure if the above derivation is correct. If I'm wrong, what's the correct mathematical derivation?