Official version v0.1

README.md

@@ -2,12 +2,44 @@
 
 The entire project is divided into two parts: `text_or_gloss2pose` and `pose2video`. "Pose" here refers to the skeleton pose video after drawing.
 
-- In the text2pose section, I first used the DWpose tool to extract the npz format pose information for each word video frame by frame, and then merged each frame's npz into a large npz file. That is to say, we know the consecutive poses corresponding to each word. These consecutive poses are stored in the merged npz file. (We can also use deep learning models to generate these poses, but for commercial-level accuracy, this original data Dict method will be much more accurate.) 
+- In the text2pose section (`Expected 4 seconds / one video`), I first used the DWpose tool to extract the npz format pose information for each word video frame by frame, and then merged each frame's npz into a large npz file. That is to say, we know the consecutive poses corresponding to each word. These consecutive poses are stored in the merged npz file. (We can also use deep learning models to generate these poses, but for commercial-level accuracy, this original data Dict method will be much more accurate.) 
 
-- In the pose2video section, we render the visualized poses of the skeletons generated by each word. I used the ControlNext-SVD model for this. 
+- In the pose2video section (`Expected 4 minutes / one video`), we render the visualized poses of the skeletons generated by each word. I used the ControlNext-SVD model for this. 
 
 # For Users
 
+1. cd StableSigner
+2. conda env create -f env/environment02_controlnext.yml
+3. run `conda activate controlnext`
+4. run `pip install -r env/requirements02_controlnext.txt`
+
+first, change `good morning afternoon`, and run:
+
+```python
+python text2pose.py --gloss "good morning afternoon" --npz-interpolation 20 --width 512 --height 768 --fps 30
+```
+
+second, change `generated_pose_video/GOOD_MORNING_AFTERNOON_25073_36863_01435_20250803_233215.mp4` & `ref_img_dict/03.jpeg`, and run:
+
+```python
+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/GOOD_MORNING_AFTERNOON_25073_36863_01435_20250803_233215.mp4 \
+  --ref_image_path ref_img_dict/03.jpeg \
+  --precision fp16 \
+  --enable_xformers \
+  --decode_chunk_size 2
+```
 
 
 # For developers
@@ -71,7 +103,7 @@ 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/GOOD_AFTERNOON_25073_01435_20250803_222800.mp4 \
+  --pose_video_path generated_pose_video/GOOD_MORNING_AFTERNOON_25073_36863_01435_20250803_233215.mp4 \
   --ref_image_path ref_img_dict/03.jpeg \
   --precision fp16 \
   --enable_xformers \

generated_pose_video/{GOOD_AFTERNOON_25073_01435_20250803_222800.mp4 → GOOD_MORNING_AFTERNOON_25073_36863_01435_20250803_233215.mp4}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c09157720812d8c72b29fd76b701f60434ba1e5299aaec2054aa49f5fd97495a
-size 1003827
+oid sha256:0fe3c60fcfffa5c18867d43d23c0c72ee4dfa7abe59b4e80d79de87ba2cffce5
+size 1685778

generated_pose_video/HELLO_WORLD_27184_63831_20250803_165003_fade.mp4

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:94919506b7131e8313c931608196190092d523574f8858d77abaf384509e6717
-size 1667961

generated_sign_video/GOOD_AFTERNOON_25073_01435_20250803_222800.mp4

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:03ae3f09974b40e1aa8e74feb86b8712060537b8eb69bd24344b5104f4d406fe
-size 585379

generated_sign_video/GOOD_AFTERNOON_25073_01435_20250803_222800_combined.mp4

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:04d3f48926a2e3b26889f47d176d7a68c157bdb8f17ae9800b19146e3bda3507
-size 815754

generated_pose_video/GOOD_MORNING_AFTERNOON_25073_36863_01435_20250803_231751.mp4 → generated_sign_video/GOOD_MORNING_AFTERNOON_25073_36863_01435_20250803_233215.mp4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:761d087c399b83e83d0246c6eb84d5cefe7290a01ceb59f1bbab2a29e7a4bf26
-size 1688288
+oid sha256:c63044ffcffb6d3b4882da43bffe6f61ff3b521c5ba5a1b9cc446a910f574e07
+size 1029963

generated_pose_video/HELLO_27184_20250803_164908.mp4 → generated_sign_video/GOOD_MORNING_AFTERNOON_25073_36863_01435_20250803_233215_combined.mp4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6351055b804acc5473154d575f65eab0946de032a266eb765a92312b8c0e8f12
-size 561551
+oid sha256:668ed7a6757f8b029cb1c66745e0d580ddceb1547d639dd21bdf5809433f6c7e
+size 1401238

utils/npz_interpolation.py

@@ -64,18 +64,42 @@ def catmull_rom_spline(p0, p1, p2, p3, num_points, tension=0.5):
     """
     points = []
     
+    # Check for invalid points (coordinates that are 0, -1, or out of reasonable bounds)
+    def is_valid_point(p):
+        if isinstance(p, np.ndarray):
+            # Check if any coordinate is -1, 0, or unreasonably large
+            return np.all(p > 0) and np.all(p < 2000)  # Assuming 2000 is max reasonable coordinate
+        return False
+    
+    # If endpoints are invalid, fall back to linear interpolation
+    if not is_valid_point(p1) or not is_valid_point(p2):
+        # Simple linear interpolation for invalid points
+        for i in range(num_points):
+            t = i / (num_points - 1)
+            point = p1 * (1 - t) + p2 * t
+            points.append(point)
+        return np.array(points)
+    
     for i in range(num_points):
         t = i / (num_points - 1)
         t2 = t * t
         t3 = t2 * t
         
-        # Catmull-Rom basis functions
+        # Catmull-Rom basis functions with clamping to prevent overshoot
         v0 = -tension * t + 2 * tension * t2 - tension * t3
         v1 = 1 + (tension - 3) * t2 + (2 - tension) * t3
         v2 = tension * t + (3 - 2 * tension) * t2 + (tension - 2) * t3
         v3 = -tension * t2 + tension * t3
         
-        point = v0 * p0 + v1 * p1 + v2 * p2 + v3 * p3
+        # Use valid control points or fall back to endpoints
+        p0_safe = p0 if is_valid_point(p0) else p1
+        p3_safe = p3 if is_valid_point(p3) else p2
+        
+        point = v0 * p0_safe + v1 * p1 + v2 * p2 + v3 * p3_safe
+        
+        # Clamp to reasonable bounds to prevent extreme values
+        point = np.clip(point, 0, 2000)
+        
         points.append(point)
     
     return np.array(points)
@@ -159,7 +183,7 @@ def interpolate_pose_npz(npz1_data, npz2_data, num_frames=10, method='catmull-ro
                         t_eased = ease_in_out_sine(t)
                         interpolated = data1 * (1 - t_eased) + data2 * t_eased
                     else:
-                        # For keypoints, use spline interpolation
+                        # For keypoints, use spline interpolation with validation
                         # Get control points
                         prev_key = f"frame_{prev_frame_key1.split('_')[1]}_{component}"
                         next_key = f"frame_{next_frame_key2.split('_')[1]}_{component}"
@@ -173,15 +197,30 @@ def interpolate_pose_npz(npz1_data, npz2_data, num_frames=10, method='catmull-ro
                             interpolated = np.zeros_like(data1)
                             for person_idx in range(data1.shape[0]):
                                 for joint_idx in range(data1.shape[1]):
-                                    # Interpolate each joint
-                                    points = catmull_rom_spline(
-                                        p0[person_idx, joint_idx],
-                                        p1[person_idx, joint_idx],
-                                        p2[person_idx, joint_idx],
-                                        p3[person_idx, joint_idx],
-                                        num_points=num_frames
-                                    )
-                                    interpolated[person_idx, joint_idx] = points[i]
+                                    # Check if both keypoints are valid (not -1 or 0)
+                                    kp1 = p1[person_idx, joint_idx]
+                                    kp2 = p2[person_idx, joint_idx]
+                                    
+                                    # Skip interpolation for invalid keypoints
+                                    if np.any(kp1 <= 0) or np.any(kp2 <= 0):
+                                        # Use the valid keypoint or zero
+                                        if np.any(kp1 > 0):
+                                            interpolated[person_idx, joint_idx] = kp1
+                                        elif np.any(kp2 > 0):
+                                            interpolated[person_idx, joint_idx] = kp2
+                                        else:
+                                            interpolated[person_idx, joint_idx] = 0
+                                    else:
+                                        # Only interpolate valid keypoints
+                                        points = catmull_rom_spline(
+                                            p0[person_idx, joint_idx],
+                                            p1[person_idx, joint_idx],
+                                            p2[person_idx, joint_idx],
+                                            p3[person_idx, joint_idx],
+                                            num_points=num_frames,
+                                            tension=0.3  # Lower tension for less overshoot
+                                        )
+                                        interpolated[person_idx, joint_idx] = points[i]
                         else:
                             # Single dimension data
                             t_eased = ease_in_out_sine(t)