rm fluid api (#641)

ppgan/models/generators/generator_firstorder.py

@@ -251,7 +251,7 @@ class Transform:
         theta_part_a = theta[:, :, :, :2]
         theta_part_b = theta[:, :, :, 2:]
 
-        transformed = paddle.fluid.layers.matmul(
+        transformed = paddle.matmul(
             *broadcast(theta_part_a, coordinates)) + theta_part_b  #M*p + m0
         transformed = transformed.squeeze(-1)
         if self.tps:

ppgan/models/starganv2_model.py

@@ -2,7 +2,6 @@
 # Users should be careful about adopting these functions in any commercial matters.
 # https://github.com/clovaai/stargan-v2#license
 
-from paddle.fluid.layers.nn import soft_relu
 from .base_model import BaseModel
 
 from paddle import nn

ppgan/models/wav2lip_hq_model.py

@@ -103,8 +103,7 @@ class Wav2LipModelHq(BaseModel):
         self.l1_loss = self.recon_loss(self.g, self.y)
 
         if self.disc_wt > 0.:
-            if isinstance(self.nets['netDH'], paddle.DataParallel
-                          ):  #paddle.fluid.dygraph.parallel.DataParallel)
+            if isinstance(self.nets['netDH'], paddle.DataParallel):
                 self.perceptual_loss = self.nets[
                     'netDH']._layers.perceptual_forward(self.g)
             else:
@@ -175,8 +174,7 @@ class Wav2LipModelHq(BaseModel):
             self.eval_recon_losses.append(l1loss.numpy().item())
 
             if self.disc_wt > 0.:
-                if isinstance(self.nets['netDH'], paddle.DataParallel
-                              ):  #paddle.fluid.dygraph.parallel.DataParallel)
+                if isinstance(self.nets['netDH'], paddle.DataParallel):
                     perceptual_loss = self.nets[
                         'netDH']._layers.perceptual_forward(
                             self.g).numpy().item()

ppgan/modules/first_order.py

@@ -468,7 +468,8 @@ class AntiAliasInterpolation2d(nn.Layer):
         inv_scale = 1 / self.scale
         int_inv_scale = int(inv_scale)
         assert (inv_scale == int_inv_scale)
-        #out = out[:, :, ::int_inv_scale, ::int_inv_scale]
+        # lite: fluid resize_nearest
+        # out = paddle.fluid.layers.resize_nearest(out, scale=self.scale)
+        out = out[:, :, ::int_inv_scale, ::int_inv_scale]
         # patch end
-        out = paddle.fluid.layers.resize_nearest(out, scale=self.scale)
         return out

tools/fom_infer.py

@@ -5,7 +5,6 @@ import cv2
 import imageio
 import time
 from tqdm import tqdm
-import paddle.fluid as fluid
 import os
 from functools import reduce
 import paddle
@@ -99,11 +98,11 @@ def main():
         driving_paths = [driving_path]
 
     # 创建 config
-    kp_detector_config = paddle_infer.Config(os.path.join(
-        args.model_path, "kp_detector.pdmodel"),
+    kp_detector_config = paddle_infer.Config(
+        os.path.join(args.model_path, "kp_detector.pdmodel"),
         os.path.join(args.model_path, "kp_detector.pdiparams"))
-    generator_config = paddle_infer.Config(os.path.join(
-        args.model_path, "generator.pdmodel"),
+    generator_config = paddle_infer.Config(
+        os.path.join(args.model_path, "generator.pdmodel"),
         os.path.join(args.model_path, "generator.pdiparams"))
     if args.device == "gpu":
         kp_detector_config.enable_use_gpu(100, 0)
@@ -194,7 +193,8 @@ def main():
             generator_output_handle = generator_predictor.get_output_handle(
                 generator_output_names[0])
             output_data = generator_output_handle.copy_to_cpu()
-            loss = paddle.abs(paddle.to_tensor(output_data) -
+            loss = paddle.abs(
+                paddle.to_tensor(output_data) -
                 paddle.to_tensor(driving_video[i])).mean().cpu().numpy()
             test_loss.append(loss)
             output_data = np.transpose(output_data, [0, 2, 3, 1])[0] * 255.0
@@ -210,8 +210,7 @@ def main():
                         fps=fps)
     metric_file = os.path.join(args.output_path, "metric.txt")
     log_file = open(metric_file, 'a')
-    loss_string = "Metric {}: {:.4f}".format(
-                  "l1 loss", np.mean(test_loss))
+    loss_string = "Metric {}: {:.4f}".format("l1 loss", np.mean(test_loss))
     log_file.close()