Update test code args parser

Author: yzhou359

README.md

@@ -30,7 +30,7 @@ Our method creates both non-photorealistic cartoon animations (top) and natural
 - Python environment 3.6
 ```
 conda create -n makeittalk_env python=3.6
-conda activate makeittalk
+conda activate makeittalk_env
 ```
 - ffmpeg (https://ffmpeg.org/download.html)
 ```
@@ -55,7 +55,7 @@ Download the following pre-trained models to `examples/ckpt` folder.
 
 ## Animate You Portraits!
 
-`Nature human faces / Paintings` (warping through Image-to-image translation module)
+### _Nature human faces / Paintings_ (warping through Image-to-image translation module)
 
 - crop your portrait image into size `256x256` and put it under `examples` folder with `.jpg` format. 
 Make sure the head is almost in the middle (check existing examples for a reference).
@@ -73,7 +73,7 @@ to amply lip motion (in x/y-axis direction) and head motion displacements, defau
 
 
 
-`Non-photorealistic cartoon faces` (warping through Delaunay triangulation)
+### _Non-photorealistic cartoon faces_ (warping through Delaunay triangulation)
 
 - animate one of the existing puppets
 

main_end2end.py

@@ -32,7 +32,7 @@
 
 
 parser = argparse.ArgumentParser()
-parser.add_argument('--jpg', type=str, default='examples/{}.jpg'.format(default_head_name))
+parser.add_argument('--jpg', type=str, default='{}.jpg'.format(default_head_name))
 parser.add_argument('--close_input_face_mouth', default=CLOSE_INPUT_FACE_MOUTH, action='store_true')
 
 
@@ -44,8 +44,8 @@
 parser.add_argument('--amp_lip_x', type=float, default=2.)
 parser.add_argument('--amp_lip_y', type=float, default=2.)
 parser.add_argument('--amp_pos', type=float, default=1.)
-parser.add_argument('--reuse_train_emb_list', type=list, default=[]) #  ['iWeklsXc0H8']) #['45hn7-LXDX8']) #['E_kmpT-EfOg']) #'iWeklsXc0H8', '29k8RtSUjE0', '45hn7-LXDX8',
-
+parser.add_argument('--reuse_train_emb_list', type=str, nargs='+', default=[]) #  ['iWeklsXc0H8']) #['45hn7-LXDX8']) #['E_kmpT-EfOg']) #'iWeklsXc0H8', '29k8RtSUjE0', '45hn7-LXDX8',
+# --reuse_train_emb_list 45hn7-LXDX8
 
 parser.add_argument('--add_audio_in', default=False, action='store_true')
 parser.add_argument('--comb_fan_awing', default=False, action='store_true')
@@ -71,7 +71,7 @@
 opt_parser = parser.parse_args()
 
 ''' STEP 1: preprocess input single image '''
-img =cv2.imread(opt_parser.jpg)
+img =cv2.imread('examples/' + opt_parser.jpg)
 predictor = face_alignment.FaceAlignment(face_alignment.LandmarksType._3D, device='cuda', flip_input=True)
 shapes = predictor.get_landmarks(img)
 if (not shapes or len(shapes) != 1):
@@ -83,7 +83,8 @@
     util.close_input_face_mouth(shape_3d)
 
 
-''' Additional manual adjustment to input face landmarks (slimmer lips and widers eyes) '''
+''' Additional manual adjustment to input face landmarks (slimmer lips and wider eyes) '''
+shape_3d[48:, 0] = (shape_3d[48:, 0] - np.mean(shape_3d[48:, 0])) * 0.95 + np.mean(shape_3d[48:, 0])
 shape_3d[49:54, 1] += 1.
 shape_3d[55:60, 1] -= 1.
 shape_3d[[37,38,43,44], 1] -=2
@@ -177,6 +178,6 @@
     ''' STEP 6: Imag2image translation '''
     model = Image_translation_block(opt_parser, single_test=True)
     with torch.no_grad():
-        model.single_test(jpg=img, fls=fl, filename=fls[i], prefix=opt_parser.jpg)
+        model.single_test(jpg=img, fls=fl, filename=fls[i], prefix=opt_parser.jpg.split('.')[0])
         print('finish image2image gen')
     os.remove(os.path.join('examples', fls[i]))

main_end2end_cartoon.py

@@ -37,7 +37,7 @@
 parser.add_argument('--amp_lip_x', type=float, default=2.0)
 parser.add_argument('--amp_lip_y', type=float, default=2.0)
 parser.add_argument('--amp_pos', type=float, default=0.8)
-parser.add_argument('--reuse_train_emb_list', default=[]) #  ['E_kmpT-EfOg']) #  ['E_kmpT-EfOg']) # ['45hn7-LXDX8'])
+parser.add_argument('--reuse_train_emb_list', default=['45hn7-LXDX8']) #  ['E_kmpT-EfOg']) #  ['E_kmpT-EfOg']) # ['45hn7-LXDX8'])
 
 
 parser.add_argument('--add_audio_in', default=False, action='store_true')
@@ -177,27 +177,27 @@
 
     os.remove(os.path.join('examples_cartoon', fls_names[i]))
 
-    # ==============================================
-    # Step 4 : Vector art morphing (only work in WINDOWS)
-    # ==============================================
-    warp_exe = os.path.join(os.getcwd(), 'facewarp', 'facewarp.exe')
-    import os
-
-    if (os.path.exists(os.path.join(output_dir, 'output'))):
-        shutil.rmtree(os.path.join(output_dir, 'output'))
-    os.mkdir(os.path.join(output_dir, 'output'))
-    os.chdir('{}'.format(os.path.join(output_dir, 'output')))
-    print(os.getcwd())
-
-    os.system('{} {} {} {} {} {}'.format(
-        warp_exe,
-        os.path.join('examples_cartoon', DEMO_CH+'.png'),
-        os.path.join(output_dir, 'triangulation.txt'),
-        os.path.join(output_dir, 'reference_points.txt'),
-        os.path.join(output_dir, 'warped_points.txt'),
-        # os.path.join(ROOT_DIR, 'puppets', sys.argv[6]),
-        '-novsync -dump'))
-    os.system('ffmpeg -y -r 62.5 -f image2 -i "%06d.tga" -i {} -shortest {}'.format(
-        ain,
-        os.path.join(output_dir, sys.argv[8])
-    ))
+    # # ==============================================
+    # # Step 4 : Vector art morphing (only work in WINDOWS)
+    # # ==============================================
+    # warp_exe = os.path.join(os.getcwd(), 'facewarp', 'facewarp.exe')
+    # import os
+    #
+    # if (os.path.exists(os.path.join(output_dir, 'output'))):
+    #     shutil.rmtree(os.path.join(output_dir, 'output'))
+    # os.mkdir(os.path.join(output_dir, 'output'))
+    # os.chdir('{}'.format(os.path.join(output_dir, 'output')))
+    # print(os.getcwd())
+    #
+    # os.system('{} {} {} {} {} {}'.format(
+    #     warp_exe,
+    #     os.path.join('examples_cartoon', DEMO_CH+'.png'),
+    #     os.path.join(output_dir, 'triangulation.txt'),
+    #     os.path.join(output_dir, 'reference_points.txt'),
+    #     os.path.join(output_dir, 'warped_points.txt'),
+    #     # os.path.join(ROOT_DIR, 'puppets', sys.argv[6]),
+    #     '-novsync -dump'))
+    # os.system('ffmpeg -y -r 62.5 -f image2 -i "%06d.tga" -i {} -shortest {}'.format(
+    #     ain,
+    #     os.path.join(output_dir, sys.argv[8])
+    # ))

src/approaches/train_audio2landmark.py

@@ -260,7 +260,7 @@ def __close_face_lip__(self, fl):
 
     def test(self, au_emb=None):
         with torch.no_grad():
-            self.__train_pass__(au_emb)
+            self.__train_pass__(au_emb, vis_fls=True)
 
     def __solve_inverse_lip2__(self, fl_dis_pred_pos_numpy):
         for j in range(fl_dis_pred_pos_numpy.shape[0]):