Update app.py

app.py

@@ -28,20 +28,26 @@ def cut_and_feather(img, feather):
     mask = np.zeros((h, w), dtype=np.uint8)
     results = face_mesh.process(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
     if not results.multi_face_landmarks:
-        return np.zeros_like(img)
-    lm = results.multi_face_landmarks[0]
-    pts = np.array([(int(p.x*w), int(p.y*h)) for p in lm.landmark], np.int32)
+        return np.zeros_like(img), None, None
+    pts = np.array([(int(p.x * w), int(p.y * h)) for p in results.multi_face_landmarks[0].landmark], np.int32)
     hull = cv2.convexHull(pts)
     cv2.fillConvexPoly(mask, hull, 255)
+    # bounding box
+    x, y, bw, bh = cv2.boundingRect(hull)
     # feather mask
     k = int(feather)
-    if k>0:
+    if k > 0:
         mask = cv2.GaussianBlur(mask, (k*2+1, k*2+1), 0)
-    fg = cv2.bitwise_and(img, img, mask=mask)
-    bg = np.zeros_like(img)
-    alpha = mask.astype(np.float32)/255.0
-    out = (fg.astype(np.float32) * alpha[...,None] + bg.astype(np.float32)*(1-alpha[...,None])).astype(np.uint8)
-    return out, mask
+    # extract face ROI
+    face_roi = img[y:y+bh, x:x+bw]
+    mask_roi = mask[y:y+bh, x:x+bw]
+    # apply mask
+    fg = cv2.bitwise_and(face_roi, face_roi, mask=mask_roi)
+    # prepare alpha
+    alpha = mask_roi.astype(np.float32) / 255.0
+    # composite onto transparent background same size
+    out = (fg.astype(np.float32) * alpha[..., None]).astype(np.uint8)
+    return out, mask_roi, (x, y, bw, bh)
 
 def get_landmarks(img, landmark_step=1):
     if img is None or face_mesh is None:
@@ -130,44 +136,66 @@ def process_video(video_path, ref_img, trans, res, step, feather):
     cap = cv2.VideoCapture(video_path)
     fps = cap.get(cv2.CAP_PROP_FPS) or 24
     total = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
-    # prepare modified reference
+    # prepare reference cut
     ref_bgr = cv2.cvtColor(ref_img, cv2.COLOR_RGB2BGR)
-    mod_ref, _ = cut_and_feather(ref_bgr, feather)
-    mod_ref = cv2.resize(mod_ref, (res, res))
-    # output tmp video
+    ref_cut, ref_mask, ref_box = cut_and_feather(ref_bgr, feather)
+    if ref_cut is None:
+        return None, None, None, None
+    # resize ref for morph
+    ref_morph = cv2.resize(ref_cut, (res, res))
+    # setup output video writer for original size
+    w_orig = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
+    h_orig = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
     tmp_vid = tempfile.NamedTemporaryFile(delete=False, suffix='.mp4').name
-    out_vid = cv2.VideoWriter(tmp_vid, cv2.VideoWriter_fourcc(*'mp4v'), fps, (res, res))
-    first_mod_frame = None; first_morphed = None
+    out_vid = cv2.VideoWriter(tmp_vid, cv2.VideoWriter_fourcc(*'mp4v'), fps, (w_orig, h_orig))
+    first_mod_frame = None
+    first_morphed = None
     for i in range(total):
         ret, frame = cap.read()
-        if not ret: break
-        frm_mod, mask = cut_and_feather(frame, feather)
-        frm_mod = cv2.resize(frm_mod, (res, res))
-        if i==0:
-            first_mod_frame = frm_mod.copy()
-            alpha = float(np.clip((trans+1)/2,0,1))
-            first_morphed = morph_faces(frm_mod, mod_ref, alpha, res, step)
-        mor = morph_faces(frm_mod, mod_ref, float(np.clip((trans+1)/2,0,1)), res, step)
-        out_vid.write(mor)
-    cap.release(); out_vid.release()
-    # convert back to RGB for Gradio display
-    return tmp_vid, cv2.cvtColor(first_mod_frame, cv2.COLOR_BGR2RGB), cv2.cvtColor(mod_ref, cv2.COLOR_BGR2RGB), cv2.cvtColor(first_morphed, cv2.COLOR_BGR2RGB)
+        if not ret:
+            break
+        # cut current frame face
+        cut, mask_roi, box = cut_and_feather(frame, feather)
+        if mask_roi is None:
+            out_vid.write(frame)
+            continue
+        x, y, bw, bh = box
+        # prepare for morph
+        cut_resized = cv2.resize(cut, (res, res))
+        alpha = float(np.clip((trans + 1) / 2, 0, 1))
+        morphed = morph_faces(cut_resized, ref_morph, alpha, res, step)
+        # store first outputs
+        if i == 0:
+            first_mod_frame = cut_resized.copy()
+            first_morphed = morphed.copy()
+        # composite morphed back to original frame
+        mor_back = cv2.resize(morphed, (bw, bh))
+        # normalize mask_roi
+        mask_norm = (mask_roi.astype(np.float32) / 255.0)[..., None]
+        # blend
+        region = frame[y:y+bh, x:x+bw].astype(np.float32)
+        blended = region * (1 - mask_norm) + mor_back.astype(np.float32) * mask_norm
+        frame[y:y+bh, x:x+bw] = blended.astype(np.uint8)
+        out_vid.write(frame)
+    cap.release()
+    out_vid.release()
+    # convert for Gradio
+    return tmp_vid, cv2.cvtColor(first_mod_frame, cv2.COLOR_BGR2RGB), cv2.cvtColor(ref_cut, cv2.COLOR_BGR2RGB), cv2.cvtColor(first_morphed, cv2.COLOR_BGR2RGB)
 
 
 # --- Gradio App ---
 css = """video, img { object-fit: contain !important; }"""
 with gr.Blocks(css=css) as iface:
-    gr.Markdown("# Enhanced Face Morph with Feathering")
+    gr.Markdown("# Enhanced Face Morph with Composite Placement")
     with gr.Row():
         vid = gr.Video(label='Input Video')
         ref = gr.Image(type='numpy', label='Reference Face Image')
     with gr.Row():
-        res = gr.Dropdown([256,384,512,768], value=512, label='Resolution')
-        step = gr.Slider(1,4, value=1, step=1, label='Landmark Sub-sampling')
-        feather = gr.Slider(0,50, value=10, step=1, label='Feather Radius')
-    trans = gr.Slider(-1.0,1.0, value=0.0, step=0.05, label='Transition Level')
+        res = gr.Dropdown([256, 384, 512, 768], value=512, label='Morph Resolution')
+        step = gr.Slider(1, 4, value=1, step=1, label='Landmark Sub-sampling')
+        feather = gr.Slider(0, 50, value=10, step=1, label='Feather Radius')
+    trans = gr.Slider(-1.0, 1.0, value=0.0, step=0.05, label='Transition Level')
     btn = gr.Button('Generate Morph 🚀')
-    progress = gr.Progress()
     out_vid = gr.Video(label='Morphed Video')
     out_mod_frame = gr.Image(label='Modified Frame[0]')
     out_mod_ref = gr.Image(label='Modified Reference')
@@ -179,8 +207,9 @@ with gr.Blocks(css=css) as iface:
         outputs=[out_vid, out_mod_frame, out_mod_ref, out_morph0],
         show_progress=True
     )
-    gr.Markdown("---\n*Workflow: cut, feather, morph, composite.*")
+    gr.Markdown("---\n*Composite the morphed face back onto the original video.*")
 
-if __name__=='__main__':
+if __name__ == '__main__':
     iface.launch(debug=True)
 
+