quick update flow

tools/scoring/aesthetic/inference.py

@@ -1,6 +1,6 @@
 # adapted from https://github.com/christophschuhmann/improved-aesthetic-predictor/blob/main/simple_inference.py
-import os
 import argparse
+import os
 from datetime import timedelta
 
 import clip
@@ -10,9 +10,9 @@ import torch
 import torch.distributed as dist
 import torch.nn as nn
 import torch.nn.functional as F
-from torch.utils.data import DataLoader, DistributedSampler
 from einops import rearrange
 from PIL import Image
+from torch.utils.data import DataLoader, DistributedSampler
 from torchvision.datasets.folder import pil_loader
 from tqdm import tqdm
 
@@ -169,6 +169,7 @@ def main():
     dist.barrier()  
 
     gathered_list = [None] * dist.get_world_size()
+    breakpoint()
     dist.all_gather_object(gathered_list, (indices_list, scores_list))
     if dist.get_rank() == 0:
         meta_new = merge_scores(gathered_list, dataset.meta, column='aes')

tools/scoring/optical_flow/inference.py

@@ -1,4 +1,4 @@
-import cv2
+import cv2  # isort:skip
 
 import argparse
 import os
@@ -17,6 +17,8 @@ from tqdm import tqdm
 from tools.datasets.utils import extract_frames
 from tools.scoring.optical_flow.unimatch import UniMatch
 
+# torch.backends.cudnn.enabled = False # This line enables large batch, but the speed is similar
+
 
 def merge_scores(gathered_list: list, meta: pd.DataFrame, column):
     # reorder
@@ -150,8 +152,8 @@ def main():
         scores_list.extend(flow_scores)
 
     # jun 3 quickfix
-    meta_local = merge_scores([(indices_list, scores_list)], dataset.meta, column='flow')
-    out_path_local = out_path.replace('.csv', f'_part_{dist.get_rank()}.csv')
+    meta_local = merge_scores([(indices_list, scores_list)], dataset.meta, column="flow")
+    out_path_local = out_path.replace(".csv", f"_part_{dist.get_rank()}.csv")
     meta_local.to_csv(out_path_local, index=False)
 
     # wait for all ranks to finish data processing