got nan in tensor.reverse()

When I run the following code, the second parameter of pixel2cam contains nan

cam_coords = pixel2cam(depth.squeeze(1), intrinsics.inverse()) # [B,3,H,W]

then, i added this code before pixel2cam

    if torch.any(torch.isnan(intrinsics.inverse())):
        print(intrinsics)
        print(intrinsics.inverse())
        print(torch.isnan(intrinsics.inverse()))
        print(torch.any(torch.isnan(intrinsics.inverse())))

but got output as follow image:


This result looks very contradictory. After entering the if torch.any(torch.isnan(intrinsics.inverse())) , why thre output of torch.any(torch.isnan(intrinsics.inverse()) is still False?