resize image before gen pose

main.py

@@ -216,7 +216,7 @@ def get_pose(img, dwpose, outfile, crop=False):
     #pil_image = Image.open("imgs/"+img).convert("RGB")
     #skeleton = dwpose(pil_image, output_type="np", include_hands=True, include_face=False)
 
-    #img.thumbnail((512,512))
+    img.thumbnail((512,512))
     out_img, pose = dwpose(img, include_hands=True, include_face=False)
 
     #print(pose['bodies'])
@@ -299,6 +299,9 @@ def prepare_inputs_inference(in_img, in_vid, frames, fps, dwpose, rembg_session,
     print(frames)
     if in_vid:
         frames = extract_frames(in_vid, fps)
+    for f in frames:
+        f.thumbnail((512,512))
+    
     print("remove background", bg_remove)
     if bg_remove:
         in_img = removebg(in_img, rembg_session)
@@ -1180,7 +1183,7 @@ def run_inference_impl(images, video_path, frames, train_steps=100, inference_st
     gc.collect()
     torch.cuda.empty_cache()
 
-    return out_vid+'.webm', results, getThumbnails(results), target_poses_coords, getThumbnails(orig_frames)
+    return out_vid+'.webm', results, getThumbnails(results), target_poses_coords, orig_frames
 
 def run_inference(images, video_path, frames, train_steps=100, inference_steps=10, fps=12, modelId="fine_tuned_pcdms", img_width=1920, img_height=1080, bg_remove=True, resize_inputs=True):
     return run_inference_impl(images, video_path, frames, train_steps, inference_steps, fps, modelId, img_width, img_height, bg_remove, resize_inputs)