fix out length exceeds than pose frames

.gitignore

@@ -1,4 +1,11 @@
 __pycache__/
 pretrained_weights/
 output/
-.venv/
+.venv/
+mlruns/
+data/
+
+*.pth
+*.pt
+*.pkl
+*.bin

app.py

@@ -114,29 +114,33 @@ class AnimateController:
         src_fps = get_fps(pose_video_path)
 
         pose_list = []
-        pose_tensor_list = []
-        pose_transform = transforms.Compose(
-            [transforms.Resize((height, width)), transforms.ToTensor()]
-        )
-        for pose_image_pil in pose_images[:length]:
+        total_length = min(length, len(pose_images))
+        for pose_image_pil in pose_images[:total_length]:
             pose_list.append(pose_image_pil)
-            pose_tensor_list.append(pose_transform(pose_image_pil))
 
         video = self.pipeline(
             ref_image,
             pose_list,
             width=width,
             height=height,
-            video_length=length,
+            video_length=total_length,
             num_inference_steps=num_inference_steps,
             guidance_scale=cfg,
             generator=generator,
         ).videos
 
+        new_h, new_w = video.shape[-2:]
+        pose_transform = transforms.Compose(
+            [transforms.Resize((new_h, new_w)), transforms.ToTensor()]
+        )
+        pose_tensor_list = []
+        for pose_image_pil in pose_images[:total_length]:
+            pose_tensor_list.append(pose_transform(pose_image_pil))
+
         ref_image_tensor = pose_transform(ref_image)  # (c, h, w)
         ref_image_tensor = ref_image_tensor.unsqueeze(1).unsqueeze(0)  # (1, c, 1, h, w)
         ref_image_tensor = repeat(
-            ref_image_tensor, "b c f h w -> b c (repeat f) h w", repeat=length
+            ref_image_tensor, "b c f h w -> b c (repeat f) h w", repeat=total_length
         )
         pose_tensor = torch.stack(pose_tensor_list, dim=0)  # (f, c, h, w)
         pose_tensor = pose_tensor.transpose(0, 1)