add fid for style gan (#347)

* add fid for style gan

* add fid for style gan

* add fid for style gan

configs/stylegan_v2_256_ffhq.yaml

@@ -23,6 +23,7 @@ model:
   params:
     gen_iters: 4
     disc_iters: 16
+  max_eval_steps: 50000
 
 export_model:
   - {name: 'gen', inputs_num: 2}
@@ -72,3 +73,11 @@ log_config:
 
 snapshot_config:
   interval: 5000
+
+validate:
+  interval: 5000
+  save_imig: False
+  metrics:
+    fid: # metric name, can be arbitrary
+      name: FID
+      batch_size: 4

ppgan/metrics/fid.py

@@ -48,7 +48,7 @@ class FID(paddle.metric.Metric):
         self.premodel_path = premodel_path
         if model is None:
             block_idx = InceptionV3.BLOCK_INDEX_BY_DIM[dims]
-            model = InceptionV3([block_idx])
+            model = InceptionV3([block_idx], normalize_input=False)
         if premodel_path is None:
             premodel_path = get_weights_path_from_url(INCEPTIONV3_WEIGHT_URL)
         self.model = model
@@ -63,18 +63,15 @@ class FID(paddle.metric.Metric):
         self.results = []
 
     def update(self, preds, gts):
-        if len(preds.shape) >=4:
-            self.preds.append(preds)
-            self.gts.append(gts)
-        else:
-            for i in range(preds.shape[0]):
-                self.preds.append(preds[i,:,:,:,:])
-                self.gts.append(gts[i,:,:,:,:])
-        
+        preds_inception, gts_inception = calculate_inception_val(
+            preds, gts, self.batch_size, self.model, self.use_GPU, self.dims)
+        self.preds.append(preds_inception)
+        self.gts.append(gts_inception)
+
     def accumulate(self):
-        self.preds = paddle.concat(self.preds, axis=0)
-        self.gts = paddle.concat(self.gts, axis=0)
-        value = calculate_fid_given_img(self.preds, self.gts, self.batch_size, self.model, self.use_GPU, self.dims)
+        self.preds = np.concatenate(self.preds, axis=0)
+        self.gts = np.concatenate(self.gts, axis=0)
+        value = calculate_fid_given_img(self.preds, self.gts)
         self.reset() 
         return value
 
@@ -82,8 +79,6 @@ class FID(paddle.metric.Metric):
         return 'FID'
 
 
-
-
 def _calculate_frechet_distance(mu1, sigma1, mu2, sigma2, eps=1e-6):
     m1 = np.atleast_1d(mu1)
     m2 = np.atleast_1d(mu2)
@@ -111,7 +106,6 @@ def _calculate_frechet_distance(mu1, sigma1, mu2, sigma2, eps=1e-6):
             m = np.max(np.abs(covmean.imag))
             raise ValueError('Imaginary component {}'.format(m))
         covmean = covmean.real
-
     tr_covmean = np.trace(covmean)
 
     return (diff.dot(diff) + np.trace(sigma1) + np.trace(sigma2) -
@@ -132,32 +126,32 @@ def _get_activations_from_ims(img, model, batch_size, dims, use_gpu):
         images = img[start:end]
         if images.shape[1] != 3:
             images = images.transpose((0, 3, 1, 2))
-
+        
         images = paddle.to_tensor(images)
         pred = model(images)[0][0]
         pred_arr[start:end] = pred.reshape([end - start, -1]).cpu().numpy()
     return pred_arr
 
 
-def _compute_statistic_of_img(img, model, batch_size, dims, use_gpu):
-    act = _get_activations_from_ims(img, model, batch_size, dims, use_gpu)
+def _compute_statistic_of_img(act):
     mu = np.mean(act, axis=0)
     sigma = np.cov(act, rowvar=False)
     return mu, sigma
 
-
-def calculate_fid_given_img(img_fake,
+def calculate_inception_val(img_fake,
                             img_real,
                             batch_size,
                             model,
                             use_gpu = True,
                             dims = 2048):
+    act_fake = _get_activations_from_ims(img_fake, model, batch_size, dims, use_gpu)
+    act_real = _get_activations_from_ims(img_real, model, batch_size, dims, use_gpu)
+    return act_fake, act_real
 
-    m1, s1 = _compute_statistic_of_img(img_fake, model, batch_size, dims,
-                                       use_gpu)
-    m2, s2 = _compute_statistic_of_img(img_real, model, batch_size, dims,
-                                       use_gpu)
+def calculate_fid_given_img(act_fake, act_real):
 
+    m1, s1 = _compute_statistic_of_img(act_fake)
+    m2, s2 = _compute_statistic_of_img(act_real)
     fid_value = _calculate_frechet_distance(m1, s1, m2, s2)
     return fid_value
 
@@ -305,3 +299,4 @@ def calculate_fid_given_paths(paths,
 
     fid_value = _calculate_frechet_distance(m1, s1, m2, s2)
     return fid_value
+

ppgan/models/styleganv2_model.py

@@ -79,7 +79,8 @@ class StyleGAN2Model(BaseModel):
                  r1_reg_weight=10.,
                  path_reg_weight=2.,
                  path_batch_shrink=2.,
-                 params=None):
+                 params=None,
+                 max_eval_steps=50000):
         """Initialize the CycleGAN class.
 
         Args:
@@ -107,6 +108,7 @@ class StyleGAN2Model(BaseModel):
         self.mean_path_length = 0
 
         self.nets['gen'] = build_generator(generator)
+        self.max_eval_steps = max_eval_steps
 
         # define discriminators
         if discriminator:
@@ -280,3 +282,15 @@ class StyleGAN2Model(BaseModel):
             self.visual_items['fake_img_ema'] = sample
 
         self.current_iter += 1
+
+    def test_iter(self, metrics=None):
+        self.nets['gen_ema'].eval()
+        batch = self.real_img.shape[0]
+        noises = [paddle.randn([batch, self.num_style_feat])]
+        fake_img, _ = self.nets['gen_ema'](noises)
+        with paddle.no_grad():
+            if metrics is not None:
+                for metric in metrics.values():
+                    metric.update(fake_img, self.real_img)
+        self.nets['gen_ema'].train()
+