add SSIM

reflow/solver.py

@@ -8,6 +8,7 @@
 from torch import autocast
 from torch.cuda.amp import GradScaler
 from nsf_hifigan.nvSTFT import STFT
+from reflow.ssim import calculate_ssim
 
 def calculate_mel_snr(gt_mel, pred_mel):
     # 计算误差图像
@@ -47,6 +48,15 @@ def calculate_mel_psnr(gt_mel, pred_mel):
     psnr = 10 * torch.log10(max_power / mse)
     return psnr
 
+def calculate_mel_ssim(gt_mel, pred_mel):
+    # B × M × T
+    pred_mel = pred_mel.transpose(-1, -2)
+    pred_mel = pred_mel[:, None]
+    gt_mel = gt_mel.transpose(-1, -2)
+    gt_mel = gt_mel[:, None]
+    ssim = calculate_ssim(pred_mel, gt_mel, size_average=True)
+    return ssim
+
 def test(args, model, vocoder, loader_test, saver):
     print(' [*] testing...')
     model.eval()
@@ -61,6 +71,7 @@ def test(args, model, vocoder, loader_test, saver):
     mel_val_snr_all = 0
     mel_val_psnr_all = 0
     mel_val_sisnr_all = 0
+    mel_val_ssim_all = 0
 
     # intialization
     num_batches = len(loader_test)
@@ -157,6 +168,7 @@ def test(args, model, vocoder, loader_test, saver):
             mel_val_snr_all += calculate_mel_snr(gt_mel_norm, pre_mel_norm).detach().cpu().numpy()
             mel_val_psnr_all += calculate_mel_psnr(gt_mel_norm, pre_mel_norm).detach().cpu().numpy()
             mel_val_sisnr_all += calculate_mel_si_snr(gt_mel_norm, pre_mel_norm).detach().cpu().numpy()
+            mel_val_ssim_all += calculate_mel_ssim(gt_mel_norm, pre_mel_norm).detach().cpu().numpy()
             mel_val_mse_all_num += 1
 
     # report
@@ -166,6 +178,7 @@ def test(args, model, vocoder, loader_test, saver):
     mel_val_snr_all /= mel_val_mse_all_num
     mel_val_psnr_all /= mel_val_mse_all_num
     mel_val_sisnr_all /= mel_val_mse_all_num
+    mel_val_ssim_all /= mel_val_mse_all_num
 
     # check
     print(' [test_ddsp_loss] test_ddsp_loss:', test_ddsp_loss)
@@ -187,6 +200,10 @@ def test(args, model, vocoder, loader_test, saver):
     saver.log_value({
         'validation/mel_val_sisnr': mel_val_sisnr_all
     })
+    print(' Mel Val SSIM', mel_val_ssim_all)
+    saver.log_value({
+        'validation/mel_val_ssim': mel_val_ssim_all
+    })
     return test_ddsp_loss, test_reflow_loss
 
 

reflow/ssim.py

@@ -0,0 +1,58 @@
+"""
+Adapted from https://github.com/Po-Hsun-Su/pytorch-ssim
+"""
+
+import torch
+import torch.nn.functional as F
+from torch.autograd import Variable
+import numpy as np
+from math import exp
+
+
+window = None
+
+
+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 _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)
+
+
+def calculate_ssim(img1, img2, window_size=11, size_average=True):
+    (_, channel, _, _) = img1.size()
+    global window
+    if window is None:
+        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)