Open-Sora: Democratizing Efficient Video Production for All
Python
quick update flow
zhengzangw0ec7a99dcf08
diff --git a/tools/scoring/aesthetic/inference.py b/tools/scoring/aesthetic/inference.pyindex fa583b1..a291914 100644--- a/tools/scoring/aesthetic/inference.py+++ b/tools/scoring/aesthetic/inference.py@@ -1,6 +1,6 @@# adapted from https://github.com/christophschuhmann/improved-aesthetic-predictor/blob/main/simple_inference.py-import osimport argparse+import osfrom datetime import timedeltaimport clip@@ -10,9 +10,9 @@ import torchimport torch.distributed as distimport torch.nn as nnimport torch.nn.functional as F-from torch.utils.data import DataLoader, DistributedSamplerfrom einops import rearrangefrom PIL import Image+from torch.utils.data import DataLoader, DistributedSamplerfrom torchvision.datasets.folder import pil_loaderfrom 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')diff --git a/tools/scoring/optical_flow/inference.py b/tools/scoring/optical_flow/inference.pyindex 7437d98..3f7d85f 100644--- a/tools/scoring/optical_flow/inference.py+++ b/tools/scoring/optical_flow/inference.py@@ -1,4 +1,4 @@-import cv2+import cv2 # isort:skipimport argparseimport os@@ -17,6 +17,8 @@ from tqdm import tqdmfrom tools.datasets.utils import extract_framesfrom 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