Add verified posture setting

ComfyUI-ControlNeXt-SVD

@@ -0,0 +1 @@
+Subproject commit 6227d71b16602c55a159316a0f72b0b4bf281e7f

README.md

@@ -16,11 +16,11 @@ The entire project is divided into two parts: `text_or_gloss2pose` and `pose2vid
 first, change `good morning afternoon`, and run:
 
 ```python
-python text2pose.py --gloss "where is my cat" --scale-y 1.4 --scale-x 1.3 --npz-interpolation 20 --width 512 --height 768 --fps 30 --hide-torso-lines false
+python text2pose.py --gloss "good morning afternoon" --scale-y 1.3 --scale-x 1.2 --npz-interpolation 20 --width 512 --height 768 --fps 30 --hide-torso-lines true --ref-image-path ref_img_dict/01.jpeg
 
 or
 
-python text2pose.py --gloss "good morning afternoon" --npz-interpolation 20 --width 512 --height 768 --fps 30
+python text2pose.py --gloss "where is my cat" --scale-y 1.4 --scale-x 1.3 --npz-interpolation 20 --width 512 --height 768 --fps 30 --hide-torso-lines false
 ```
 
 second, change `generated_pose_video/GOOD_MORNING_AFTERNOON_25073_36863_01435_20250804_160832.mp4` & `ref_img_dict/03.jpeg`, and run:
@@ -190,7 +190,8 @@ CUDA_VISIBLE_DEVICES=0 python ControlNeXt-SVD-v2/run_controlnext_optimized.py \
   --pose_video_path generated_pose_video/GOOD_MORNING_AFTERNOON_25073_36863_01435_20250804_160832.mp4 \
   --ref_image_path ref_img_dict/03.jpeg \
   --precision fp16 \
-  --enable_xformers \
-  --decode_chunk_size 2
+  --decode_chunk_size 2 \
+  --enable_geglu_optimization \
+  --enable_xformers 
 ```
 

generated_pose_video/{WHERE_IS_MY_CAT_63084_44900_37469_65313_20250804_195315.mp4 → GOOD_MORNING_AFTERNOON_25073_36863_01435_20250804_214708.mp4}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2c838110d6b0c453e18eb6b62aebeca205d5a1e7e32ddd35c042daa6e3ef7fed
-size 2813937
+oid sha256:e0d25a5c9af662adc2c417bbd9564cb71287f58ae6ef92cbdcc71dccd6ab6966
+size 1668013

generated_sign_video/WHERE_IS_MY_CAT_63084_44900_37469_65313_20250804_195315.mp4 → generated_pose_video/HELLO_WORLD_27184_63831_20250804_215328.mp4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:771c9a92c0901a0fc897c5a2c7142954ba7dfc02cde76834b4c8e678fa532bb9
-size 3010816
+oid sha256:2ff6dbbc50441746c7d494956bc72987878f359dd5ff96c765adb51e5631becb
+size 2183071

generated_sign_video/WHERE_IS_MY_CAT_63084_44900_37469_65313_20250804_195315_combined.mp4 → generated_pose_video/I_WILL_ATTEND_THIS_DINNER_PARTY_28794_63364_04007_66628_16327_41325_20250804_215724.mp4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:51601f233e5df4c9888351074699c398c934f7041dafb1f18e084e315fd13e86
-size 3590703
+oid sha256:ad2836123b570c7aa716e4f02e447e1e004a3d439fbcc9a6e282175475d4539c
+size 4409427

generated_pose_video/NOT_FIND_YET_38868_21853_18537_20250804_220326.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:912d9a76db09fe053361692239ba9644a47f144ec7832a2e38850811b9e5b652
+size 2095895

generated_pose_video/NOT_FIND_YET_38868_21853_38944_20250804_223748.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23b47ae5720cb5d7ba9b5e34a63e6f31f1c136d440a6a43c9113ad8a5ff5982c
+size 1340578

generated_sign_video/GOOD_MORNING_AFTERNOON_25073_36863_01435_20250804_214708.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:11c9bec1c44c66b4e8a914f4cf3f08bdfbc0f0ed7a476c6aa73cc8129c6df552
+size 1669008

generated_sign_video/GOOD_MORNING_AFTERNOON_25073_36863_01435_20250804_214708_combined.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9601bc572dc3c58fe6bdb055b1254c580fb7f33642e3313d81b06598a349d980
+size 2054929

text2pose.py

@@ -32,6 +32,7 @@ class SignLanguageQA:
     def __init__(self):
         self.gloss_data = None
         self.load_json_mapping()
+        self.detector = None  # Lazy load DWpose detector when needed
 
     def load_json_mapping(self):
         """Load JSON mapping file"""
@@ -209,13 +210,17 @@ class SignLanguageQA:
         frame_keys = [k for k in npz_data.keys() if k.startswith('frame_') and k.endswith('_bodies')]
         return len(frame_keys)
 
-    def create_pose_frame(self, frame_data, width=480, height=480, scale_x=1.0, scale_y=1.0, hide_torso_lines=False):
-        """Create single frame pose image with progressive scaling"""
+    def create_pose_frame(self, frame_data, width=480, height=480, scale_x=1.0, scale_y=1.0, hide_torso_lines=False, y_offset=0):
+        """Create single frame pose image with progressive scaling and optional y offset"""
         try:
             # Apply progressive scaling if needed
             if scale_x != 1.0 or scale_y != 1.0:
                 frame_data = self.apply_progressive_scaling(frame_data, scale_x, scale_y)
             
+            # Apply y offset if provided
+            if y_offset != 0:
+                frame_data = self.apply_y_offset(frame_data, y_offset)
+            
             processed_pred = process_pose_data(frame_data, height, width)
 
             vis_img = draw_pose(
@@ -385,14 +390,136 @@ class SignLanguageQA:
                     scaled_data['faces'] = faces
         
         return scaled_data
+    
+    def apply_y_offset(self, frame_data, y_offset):
+        """Apply vertical offset to all pose points"""
+        offset_data = copy.deepcopy(frame_data)
+        
+        # Apply offset to bodies
+        if 'bodies' in offset_data and offset_data['bodies'] is not None:
+            offset_data['bodies'][:, 1] += y_offset
+        
+        # Apply offset to hands
+        if 'hands' in offset_data and offset_data['hands'] is not None:
+            if len(offset_data['hands'].shape) == 3:
+                offset_data['hands'][:, :, 1] += y_offset
+            else:
+                offset_data['hands'][:, 1] += y_offset
+        
+        # Apply offset to faces
+        if 'faces' in offset_data and offset_data['faces'] is not None:
+            if len(offset_data['faces'].shape) == 3:
+                offset_data['faces'][:, :, 1] += y_offset
+            else:
+                offset_data['faces'][:, 1] += y_offset
+        
+        return offset_data
+    
+    def get_reference_head_top(self, ref_image_path, width, height):
+        """Get the top of head position from reference image"""
+        try:
+            # Lazy load detector
+            if self.detector is None:
+                device = "cuda:0" if torch.cuda.is_available() else "cpu"
+                self.detector = DWposeDetector(device=device)
+            
+            # Load and detect pose in reference image
+            ref_image = Image.open(ref_image_path).convert("RGB")
+            
+            # Resize reference image to match target dimensions
+            ref_image = ref_image.resize((width, height), Image.LANCZOS)
+            
+            # Detect pose
+            pose_data = self.detector(ref_image, draw_pose=False)
+            
+            # Find head top from face points
+            if 'faces' in pose_data and pose_data['faces'] is not None and len(pose_data['faces']) > 0:
+                faces = pose_data['faces']
+                if len(faces.shape) == 3:
+                    # Multiple faces, use first one
+                    face_points = faces[0]
+                else:
+                    face_points = faces
+                
+                # Get all valid Y coordinates
+                y_coords = face_points[:, 1]
+                valid_y = y_coords[y_coords > 0]
+                
+                if len(valid_y) > 0:
+                    # Head top is the minimum Y coordinate (top of the image)
+                    head_top_y = np.min(valid_y)
+                    return head_top_y
+            
+            # If no face found, try to use body points
+            if 'bodies' in pose_data and pose_data['bodies'] is not None:
+                bodies = pose_data['bodies']
+                # Look for head-related points (eyes, ears, nose)
+                # Indices 0=nose, 14=right eye, 15=left eye, 16=right ear, 17=left ear
+                head_indices = [0, 14, 15, 16, 17]
+                head_y_coords = []
+                
+                for idx in head_indices:
+                    if idx < len(bodies) and bodies[idx, 1] > 0:
+                        head_y_coords.append(bodies[idx, 1])
+                
+                if head_y_coords:
+                    return min(head_y_coords)
+            
+            return None
+            
+        except Exception as e:
+            print(f"⚠️  Failed to get reference head position: {e}")
+            return None
+    
+    def get_pose_head_top(self, frame_data):
+        """Get the top of head position from pose frame data"""
+        # Try faces first
+        if 'faces' in frame_data and frame_data['faces'] is not None:
+            faces = frame_data['faces']
+            if len(faces.shape) == 3:
+                # Multiple faces, use first one
+                face_points = faces[0]
+            else:
+                face_points = faces
+            
+            # Get all valid Y coordinates
+            y_coords = face_points[:, 1]
+            valid_y = y_coords[y_coords > 0]
+            
+            if len(valid_y) > 0:
+                return np.min(valid_y)
+        
+        # Fall back to body points
+        if 'bodies' in frame_data and frame_data['bodies'] is not None:
+            bodies = frame_data['bodies']
+            # Look for head-related points
+            head_indices = [0, 14, 15, 16, 17]  # nose, eyes, ears
+            head_y_coords = []
+            
+            for idx in head_indices:
+                if idx < len(bodies) and bodies[idx, 1] > 0:
+                    head_y_coords.append(bodies[idx, 1])
+            
+            if head_y_coords:
+                return min(head_y_coords)
+        
+        return None
 
     def generate_video_from_glosses(self, gloss_list, output_path, fps=30, width=480, height=480,
                                     smoothing_frames=5, smoothing_method='none', npz_interpolation_frames=0,
-                                    scale_x=1.0, scale_y=1.0, hide_torso_lines=False):
-        """Generate video from gloss list with optional progressive scaling"""
+                                    scale_x=1.0, scale_y=1.0, hide_torso_lines=False, ref_image_path=None):
+        """Generate video from gloss list with optional progressive scaling and head alignment"""
         # Create output directory
         output_path.parent.mkdir(parents=True, exist_ok=True)
         
+        # Get reference head position if provided
+        ref_head_top = None
+        y_offset = 0
+        if ref_image_path:
+            ref_head_top = self.get_reference_head_top(ref_image_path, width, height)
+            if ref_head_top is not None:
+                print(f"📏 Reference head top position: {ref_head_top:.2f}")
+        
         # Collect all frames from all glosses
         all_pose_frames = []
         all_npz_data = []  # Store NPZ data for interpolation
@@ -441,7 +568,15 @@ class SignLanguageQA:
                     for frame_num in range(start_frame, end_frame + 1):
                         frame_data = self.get_frame_data(npz_data, frame_num)
                         if frame_data and self.check_frame_quality(frame_data):
-                            pose_frame = self.create_pose_frame(frame_data, width, height, scale_x, scale_y, hide_torso_lines)
+                            # Calculate y offset for first frame if reference provided
+                            if ref_head_top is not None and y_offset == 0 and frame_num == start_frame:
+                                # Get head top from current pose
+                                current_head_top = self.get_pose_head_top(frame_data)
+                                if current_head_top is not None:
+                                    y_offset = ref_head_top - current_head_top
+                                    print(f"📐 Calculated Y offset: {y_offset:.2f}")
+                            
+                            pose_frame = self.create_pose_frame(frame_data, width, height, scale_x, scale_y, hide_torso_lines, y_offset)
                             if pose_frame is not None:
                                 gloss_frames.append(pose_frame)
                     
@@ -511,7 +646,7 @@ class SignLanguageQA:
                             
                             # Create visual frame
                             if frame_data:
-                                pose_frame = self.create_pose_frame(frame_data, width, height, scale_x, scale_y, hide_torso_lines)
+                                pose_frame = self.create_pose_frame(frame_data, width, height, scale_x, scale_y, hide_torso_lines, y_offset)
                                 if pose_frame is not None:
                                     interpolated_pose_frames.append(pose_frame)
                     
@@ -767,6 +902,8 @@ def main():
                         help='Progressive Y-axis scaling factor (default: 1.0)')
     parser.add_argument('--hide-torso-lines', type=str, choices=['true', 'false'], default='false',
                         help='Hide torso lines from neck to hips (set to black). Choices: true/false (default: false)')
+    parser.add_argument('--ref-image-path', type=str, default=None,
+                        help='Reference image path for head alignment')
 
     args = parser.parse_args()
 
@@ -801,7 +938,8 @@ def main():
         success = qa_system.generate_video_from_glosses(
             gloss_list, output_file, args.fps, args.width, args.height,
             args.smoothing_frames, args.smoothing, args.npz_interpolation,
-            args.scale_x, args.scale_y, args.hide_torso_lines == 'true'
+            args.scale_x, args.scale_y, args.hide_torso_lines == 'true',
+            args.ref_image_path
         )
 
         if not success:

useless/abandon_cmd.md

@@ -10,13 +10,32 @@ CUDA_VISIBLE_DEVICES=0 python ControlNeXt-SVD-v2/run_controlnext_optimized.py \
   --width 512 \
   --controlnext_path ControlNeXt-SVD-v2/pretrained/controlnet.bin \
   --unet_path ControlNeXt-SVD-v2/pretrained/unet.bin \
-  --pose_video_path generated_pose_video/HELLO_WORLD_27184_63831_20250804_163910.mp4 \
+  --pose_video_path generated_pose_video/GOOD_MORNING_AFTERNOON_25073_36863_01435_20250804_214708.mp4 \
   --ref_image_path ref_img_dict/01.jpeg \
   --precision fp16 \
   --decode_chunk_size 2 \
   --enable_geglu_optimization \
   --enable_xformers 
 
+CUDA_VISIBLE_DEVICES=0 python ControlNeXt-SVD-v2/run_controlnext_optimized.py \
+  --pretrained_model_name_or_path stabilityai/stable-video-diffusion-img2vid-xt-1-1 \
+  --output_dir generated_sign_video \
+  --max_frame_num 240 \
+  --guidance_scale 3 \
+  --batch_frames 24 \
+  --sample_stride 2 \
+  --overlap 4 \
+  --height 768 \
+  --width 512 \
+  --controlnext_path ControlNeXt-SVD-v2/pretrained/controlnet.bin \
+  --unet_path ControlNeXt-SVD-v2/pretrained/unet.bin \
+  --pose_video_path generated_pose_video/HELLO_WORLD_27184_63831_20250804_215328.mp4 \
+  --ref_image_path ref_img_dict/04.jpeg \
+  --precision fp16 \
+  --decode_chunk_size 2 \
+  --enable_geglu_optimization \
+  --enable_xformers 
+
 CUDA_VISIBLE_DEVICES=1 python ControlNeXt-SVD-v2/run_controlnext_optimized.py \
   --pretrained_model_name_or_path stabilityai/stable-video-diffusion-img2vid-xt-1-1 \
   --output_dir generated_sign_video \
@@ -29,8 +48,8 @@ CUDA_VISIBLE_DEVICES=1 python ControlNeXt-SVD-v2/run_controlnext_optimized.py \
   --width 512 \
   --controlnext_path ControlNeXt-SVD-v2/pretrained/controlnet.bin \
   --unet_path ControlNeXt-SVD-v2/pretrained/unet.bin \
-  --pose_video_path generated_pose_video/I_WILL_ATTEND_THIS_DINNER_PARTY_28794_63364_04007_66628_16327_41325_20250804_164010.mp4 \
-  --ref_image_path ref_img_dict/02.jpeg \
+  --pose_video_path generated_pose_video/code/StableSigner3/I_WILL_ATTEND_THIS_DINNER_PARTY_28794_63364_04007_66628_16327_41325_20250804_215724.mp4 \
+  --ref_image_path ref_img_dict/03.jpeg \
   --precision fp16 \
   --decode_chunk_size 2 \
   --enable_geglu_optimization \
@@ -49,7 +68,7 @@ CUDA_VISIBLE_DEVICES=2 python ControlNeXt-SVD-v2/run_controlnext_optimized.py \
   --controlnext_path ControlNeXt-SVD-v2/pretrained/controlnet.bin \
   --unet_path ControlNeXt-SVD-v2/pretrained/unet.bin \
   --pose_video_path generated_pose_video/NOT_FIND_YET_38868_21853_18535_20250804_164149.mp4 \
-  --ref_image_path ref_img_dict/dsgs_ref.jpg \
+  --ref_image_path ref_img_dict/asl_ref_v.png \
   --precision fp16 \
   --decode_chunk_size 2 \
   --enable_geglu_optimization \
@@ -67,7 +86,7 @@ CUDA_VISIBLE_DEVICES=3 python ControlNeXt-SVD-v2/run_controlnext_optimized.py \
   --width 512 \
   --controlnext_path ControlNeXt-SVD-v2/pretrained/controlnet.bin \
   --unet_path ControlNeXt-SVD-v2/pretrained/unet.bin \
-  --pose_video_path generated_pose_video/WHERE_IS_MY_CAT_63084_44900_37469_65313_20250804_195315.mp4 \
+  --pose_video_path generated_pose_video/WHERE_IS_MY_CAT_63084_66583_37469_65313_20250804_205149.mp4 \
   --ref_image_path ref_img_dict/lis_ref.jpg \
   --precision fp16 \
   --decode_chunk_size 2 \