video process resolution

inference/v2v_data/demo.py

@@ -13,7 +13,7 @@ import torch.nn.functional as F
 from transformers import Qwen2_5_VLForConditionalGeneration, AutoTokenizer, AutoProcessor
 from qwen_vl_utils import process_vision_info
 
-def get_center_crop_resolution(original_resoultion, target_aspect_ratio=(3, 4)):
+def get_center_crop_resolution(original_resoultion, height = 576, width = 1024,target_aspect_ratio=(3, 4)):
     target_h, target_w = target_aspect_ratio
     aspect_ratio = target_w / target_h
 
@@ -24,8 +24,8 @@ def get_center_crop_resolution(original_resoultion, target_aspect_ratio=(3, 4)):
         crop_w = original_w
         crop_h = int(crop_w / aspect_ratio)
 
-    resized_h = original_resoultion[0] # previous 576
-    resized_w = original_resoultion[1] # previous 1024
+    resized_h = height # previous 576
+    resized_w = width # previous 1024
     
     h_ratio = resized_h / original_h
     w_ratio = resized_w / original_w
@@ -117,7 +117,7 @@ class GetAnchorVideos:
         frame_shape = vr[0].shape  # (H, W, 3)
         ori_resolution = frame_shape[:2]
         print(f"==> original video shape: {frame_shape}")
-        target_resolution = get_center_crop_resolution(ori_resolution)
+        target_resolution = get_center_crop_resolution(original_resoultion = ori_resolution, height = opts.height, width = opts.width)
         print(f"==> target video shape resized: {target_resolution}")
 
         prompt = self.get_caption(opts, opts.video_path)