update

README.md

@@ -1,8 +1,8 @@
 ---
-title: FETE
-emoji: 👀
-colorFrom: blue
-colorTo: gray
+title: Free-View Expressive Talking Head Video Editing
+emoji: 🤖
+colorFrom: red
+colorTo: yellow
 sdk: gradio
 sdk_version: 3.41.0
 app_file: app.py

assets/coords/sample1.npz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5323c1d6ed9dbda978859aed00bd6b4ec2ca122dbc556821edd4add44181c360
-size 647
+oid sha256:5b200f395b09505d61f3efb67feaacbbd5bb358e75b476c4da083e4a7cef58af
+size 525

assets/coords/sample2.npz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aaf2424b58f2c32b3b55d172a4437bd73687ca2346e4928575c481fa024cf41d
-size 1150
+oid sha256:3ac70dd3972f406d9e8195283d11395a7b1e2528bdbdec4a3420eeac919489c9
+size 909

assets/coords/sample3.npz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fb9c284ec586ffe5cb9af9e715907810ce6f07694ef4245c29bb06646a04f3ed
-size 839
+oid sha256:246e4910d5ae9937f2d692beb6d6267dcb2f09bf7b7e0bd75d373a167289cf08
+size 598

assets/coords/sample4.npz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa3cd8b488cbd368c22b8d10f072d8a55cc73c3c0e8cea6e11420f2743cc00c4
-size 901
+oid sha256:497b14d4185a447327fac69602b66997dc791ff333ead12680c36e3e27d20195
+size 656

assets/coords/sample5.npz

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:1ee15a8dd3b47bc036a4502ccd60a0a5c29262a5581593e8f97c11e18b389e67
-size 974

attributtes_utils.py

@@ -40,7 +40,7 @@ def input_emotion(emotion_select="neutral"):
 def input_blink(blink_select="yes"):
     if blink_select == "yes":
         blink = [[1.0], [1.0], [1.0], [1.0], [1.0], [1.0], [1.0], [0.8], [0.6], [0.0], [0.0], [1.0]]
-        blink = blink + blink
+        blink = blink + blink + blink
     else:
         blink = [[1.0] for _ in range(2)]
     return blink

inference_util.py

@@ -293,7 +293,6 @@ def infenrece(model, face_path, audio_path, pose, emotion, blink, preview=False)
             y1, y2, x1, x2 = int(y1), int(y2), int(x1), int(x2)
             y = round(y2 - y1)
             x = round(x2 - x1)
-            # print(x, y, p.shape)
             p = cv2.resize(p.astype(np.uint8), (x, y))
 
             try:
@@ -301,7 +300,6 @@ def infenrece(model, face_path, audio_path, pose, emotion, blink, preview=False)
             except Exception as e:
                 print(e)
                 f[y1 : y1 + y, x1 : x1 + x] = p
-            # out.write(f[100:-20])
             f = remove_black(f)
             if preview:
                 cv2.imwrite(outfile, f, [int(cv2.IMWRITE_JPEG_QUALITY), 95])

preprocess_videos.py

@@ -66,8 +66,7 @@ def face_detect(images, pads):
         y_gap, x_gap = (y2 - y1)//2, (x2 - x1)//2
         coords_ = [y1 - y_gap, y2 + y_gap, x1 - x_gap, x2 + x_gap]
 
-        # smooth the coords
-        _, coords = get_squre_coords(coords_, image, None)
+        _, coords = get_squre_coords(coords_, image)
 
         y1, y2, x1, x2 = coords
         y1 = max(0, y1)
@@ -80,7 +79,7 @@ def face_detect(images, pads):
     print("Number of frames cropped: {}".format(len(results)))
     print("First coords: {}".format(results[0]))
     boxes = np.array(results)
-    boxes = get_smoothened_boxes(boxes, T=5)
+    boxes = get_smoothened_boxes(boxes, T=15)
     # results = [[image[y1:y2, x1:x2], (y1, y2, x1, x2)] for image, (x1, y1, x2, y2) in zip(images, boxes)]
 
     del detector