Update app.py

app.py

@@ -1,10 +1,3 @@
-import subprocess
-import os
-
-os.chdir('DenseAV')
-# subprocess.run(["cd", "DenseAV"])
-subprocess.run(["pip", "install", "-e", "."])
-
 import csv
 import os
 import tempfile
@@ -17,10 +10,11 @@ import torchvision.transforms as T
 from PIL import Image
 # from featup.util import norm
 from torchaudio.functional import resample
-
-
 from os.path import join
 
+from DenseAV.denseav.train import LitAVAligner
+from DenseAV.denseav.plotting import plot_attention_video, plot_2head_attention_video, plot_feature_video
+from DenseAV.denseav.shared import norm, crop_to_divisor, blur_dim
 
 mode = "hf"
 
@@ -76,19 +70,9 @@ video_output2 = gr.Video(label="Multi-Head Audio Video Attention (Only Availible
                          height=480)
 video_output3 = gr.Video(label="Visual Features", height=480)
 
-models = {}
-
+models = {o: LitAVAligner.from_pretrained(f"mhamilton723/DenseAV-{o}") for o in options}
 
 def process_video(video, model_option):
-    from denseav.train import LitAVAligner
-    from denseav.plotting import plot_attention_video, plot_2head_attention_video, plot_feature_video
-    from denseav.shared import norm, crop_to_divisor, blur_dim
-
-    global models
-
-    if len(models) == 0:
-        models = {o: LitAVAligner.from_pretrained(f"mhamilton723/DenseAV-{o}") for o in options}
-    
     # model = models[model_option].cuda()
     model = models[model_option]