[cleanup] fix deprecated APIs, remove dead code and unused imports

- Replace st.experimental_rerun() with st.rerun() in main_streamlit.py
- Remove dead code after return statements in mm_utils.py
- Remove unused divide_to_patches function in mm_utils.py
- Remove duplicate os imports in hevc_feature_decoder_mv.py and step1_extract_video_features.py
- Remove unused json import in inverted_index.py
- Fix np.bool deprecation warning in pack_ocr.py with proper guards
- Remove commented-out debug code in step1_extract_video_features.py

Author: Luodian

llava_next/llava/mm_utils.py

@@ -257,28 +257,6 @@ def resize_and_pad_image(image, target_resolution):
     return new_image
 
 
-def divide_to_patches(image, patch_size):
-    """
-    Divides an image into patches of a specified size.
-
-    Args:
-        image (PIL.Image.Image): The input image.
-        patch_size (int): The size of each patch.
-
-    Returns:
-        list: A list of PIL.Image.Image objects representing the patches.
-    """
-    patches = []
-    width, height = image.size
-    for i in range(0, height, patch_size):
-        for j in range(0, width, patch_size):
-            box = (j, i, j + patch_size, i + patch_size)
-            patch = image.crop(box)
-            patches.append(patch)
-
-    return patches
-
-
 def get_anyres_image_grid_shape(image_size, grid_pinpoints, patch_size):
     """
     Calculate the shape of the image patch grid after the preprocessing for images of any resolution.
@@ -352,23 +330,6 @@ def process_anyres_image(image, processor, grid_pinpoints):
         grid_thw = [1, best_resolution[1] // 14, best_resolution[0] // 14]
         return {'pixel_values': torch.cat(image_patches, dim=0), 'grid_thw': grid_thw}
 
-    patches = divide_to_patches(image_padded, processor.crop_size["height"])
-
-    # FIXME: this seems to be a bug that it resizes instead of pad.
-    # but to keep it consistent with previous, i will keep it as it is
-    # TODO: uncomment below to ablate with the padding
-    if isinstance(processor.size, dict):
-        shortest_edge = processor.size["shortest_edge"]
-    else:
-        shortest_edge = min(processor.size)
-    image_original_resize = image.resize((shortest_edge, shortest_edge))
-    # image_padded_square = expand2square(image, tuple(int(x*255) for x in processor.image_mean))
-    # image_original_resize = image_padded_square.resize((processor.size['shortest_edge'], processor.size['shortest_edge']))
-
-    image_patches = [image_original_resize] + patches
-    image_patches = [processor.preprocess(image_patch, return_tensors="pt")["pixel_values"][0] for image_patch in image_patches]
-    return torch.stack(image_patches, dim=0)
-
 
 def load_image_from_base64(image):
     return Image.open(BytesIO(base64.b64decode(image)))

tools/inverted_index.py

@@ -1,6 +1,5 @@
 import numpy as np
 import pickle
-import json
 import argparse
 from collections import defaultdict
 from pathlib import Path

tools/kmeans/step1_extract_video_features.py

@@ -378,8 +378,6 @@ def main(args):
         my_videos = all_videos
         print(f"[Rank {rank}] Processing all {len(my_videos)} videos")
 
-    # with open(f"{args.input}_padding.txt", 'w') as f:
-    #     f.writelines(all_videos)
     # Load checkpoint
     output_dir = Path(args.output)
     output_dir.mkdir(parents=True, exist_ok=True)
@@ -448,8 +446,7 @@ def main(args):
                 print(f"Final feature shape: {cleaned_features.shape}")
             elif pad_count == features_per_file:
                 # 整个最后一个文件都是 padding，直接删掉文件更合理
-                import os
-                os.remove(str(last_npy_path))
+                last_npy_path.unlink()
                 print(f"Removed entire last file since it was all padding ({pad_count} rows).")
             else:
                 # 需要继续往前删前一个文件，当前代码未实现，给出提示

tools/main_streamlit.py

@@ -462,7 +462,7 @@ def main():
                     # 保存路径到会话状态
                     st.session_state.index_file = index_file
                     st.session_state.video_list_file = video_list_file
-                    st.experimental_rerun()
+                    st.rerun()
                 else:
                     st.error("请输入所有必要的文件路径")
 
@@ -486,7 +486,7 @@ def main():
             del st.session_state.index_file
         if 'video_list_file' in st.session_state:
             del st.session_state.video_list_file
-        st.experimental_rerun()
+        st.rerun()
 
     # 配置参数
     st.sidebar.subheader("可视化配置")

tools/tools_for_hevc/hevc_feature_decoder_mv.py

@@ -703,7 +703,6 @@ def _readFrame(self):
             self.height + (self.height >> 1), self.width
         )
 
-        import os
         if int(os.environ.get('UMT_HEVC_Y_ONLY', '1')) != 0:
             y = all_yuv_data[:self.height, :self.width]
             y_res = all_yuv_data_residual[:self.height, :self.width]

tools/tools_for_ocr/pack_ocr.py

@@ -1,8 +1,13 @@
 import os
 import json
 import pathlib
+import warnings
 import numpy as np
-np.bool = np.bool_  # 解决 mxnet 与 numpy 冲突
+# mxnet uses deprecated np.bool; suppress warning and patch
+with warnings.catch_warnings():
+    warnings.filterwarnings('ignore', category=DeprecationWarning)
+    if not hasattr(np, 'bool') or np.bool is bool:
+        np.bool = np.bool_
 
 import mxnet as mx
 import mxnet.recordio as recordio