trellis/utils/loss_utils.py

import torch
import torch.nn.functional as F
from torch.autograd import Variable
from math import exp
from lpips import LPIPS


def smooth_l1_loss(pred, target, beta=1.0):
    diff = torch.abs(pred - target)
    loss = torch.where(diff < beta, 0.5 * diff ** 2 / beta, diff - 0.5 * beta)
    return loss.mean()


def l1_loss(network_output, gt):
    return torch.abs((network_output - gt)).mean()


def l2_loss(network_output, gt):
    return ((network_output - gt) ** 2).mean()


def gaussian(window_size, sigma):
    gauss = torch.Tensor([exp(-(x - window_size // 2) ** 2 / float(2 * sigma ** 2)) for x in range(window_size)])
    return gauss / gauss.sum()


def create_window(window_size, channel):
    _1D_window = gaussian(window_size, 1.5).unsqueeze(1)
    _2D_window = _1D_window.mm(_1D_window.t()).float().unsqueeze(0).unsqueeze(0)
    window = Variable(_2D_window.expand(channel, 1, window_size, window_size).contiguous())
    return window


def psnr(img1, img2, max_val=1.0):
    mse = F.mse_loss(img1, img2)
    return 20 * torch.log10(max_val / torch.sqrt(mse))


def ssim(img1, img2, window_size=11, size_average=True):
    channel = img1.size(-3)
    window = create_window(window_size, channel)

    if img1.is_cuda:
        window = window.cuda(img1.get_device())
    window = window.type_as(img1)

    return _ssim(img1, img2, window, window_size, channel, size_average)

def _ssim(img1, img2, window, window_size, channel, size_average=True):
    mu1 = F.conv2d(img1, window, padding=window_size // 2, groups=channel)
    mu2 = F.conv2d(img2, window, padding=window_size // 2, groups=channel)

    mu1_sq = mu1.pow(2)
    mu2_sq = mu2.pow(2)
    mu1_mu2 = mu1 * mu2

    sigma1_sq = F.conv2d(img1 * img1, window, padding=window_size // 2, groups=channel) - mu1_sq
    sigma2_sq = F.conv2d(img2 * img2, window, padding=window_size // 2, groups=channel) - mu2_sq
    sigma12 = F.conv2d(img1 * img2, window, padding=window_size // 2, groups=channel) - mu1_mu2

    C1 = 0.01 ** 2
    C2 = 0.03 ** 2

    ssim_map = ((2 * mu1_mu2 + C1) * (2 * sigma12 + C2)) / ((mu1_sq + mu2_sq + C1) * (sigma1_sq + sigma2_sq + C2))

    if size_average:
        return ssim_map.mean()
    else:
        return ssim_map.mean(1).mean(1).mean(1)


loss_fn_vgg = None
def lpips(img1, img2, value_range=(0, 1)):
    global loss_fn_vgg
    if loss_fn_vgg is None:
        loss_fn_vgg = LPIPS(net='vgg').cuda().eval()
    # normalize to [-1, 1]
    img1 = (img1 - value_range[0]) / (value_range[1] - value_range[0]) * 2 - 1
    img2 = (img2 - value_range[0]) / (value_range[1] - value_range[0]) * 2 - 1
    return loss_fn_vgg(img1, img2).mean()


def normal_angle(pred, gt):
    pred = pred * 2.0 - 1.0
    gt = gt * 2.0 - 1.0
    norms = pred.norm(dim=-1) * gt.norm(dim=-1)
    cos_sim = (pred * gt).sum(-1) / (norms + 1e-9)
    cos_sim = torch.clamp(cos_sim, -1.0, 1.0)
    ang = torch.rad2deg(torch.acos(cos_sim[norms > 1e-9])).mean()
    if ang.isnan():
        return -1
    return ang