Update app.py

app.py

@@ -1,10 +1,24 @@
 import gradio as gr
 import pandas as pd
-import sketch
-month=[['April 22', 207], ['May 22',209], ['June 22', 211], ['July 22', 190], ['August 22', 194], ['Sept 22', 199], ['Oct 22',187], ['Nov 22', 187], ['Dec 22', 205], ['Jan 23', 210], ['Feb 23', 210], ['March 23', 209]]
-df=pd.DataFrame(month,columns= ['Month', 'Max Demand in GW'])
-def answer(question):
-  out=df.sketch.ask(question,call_display=False)
-  return out
-demo = gr.Interface(fn=answer, inputs='text',outputs='text',examples=[['which month has the highest max demand']])
+from torchvision.io import read_video
+import torch.nn.functional as F
+import torch, hiera
+df=pd.read_csv('/content/Kinetic400.csv')
+model = hiera.hiera_base_16x224(pretrained=True, checkpoint="mae_k400_ft_k400")
+def recognize(vid):
+  frames, audio, info = read_video(vid, pts_unit='sec', output_format='THWC')
+  frames = frames.float() / 255  # Convert from byte to float
+  frames = torch.stack([frames[:64], frames[64:128]], dim=0)
+  frames = frames[:, ::4]  # Sample every 4 frames
+  frames = frames.permute(0, 4, 1, 2, 3).contiguous()
+  frames = F.interpolate(frames, size=(16, 224, 224), mode="trilinear")
+  torch.Size([2, 3, 16, 224, 224])
+  frames = frames - torch.tensor([0.45, 0.45, 0.45]).view(1, -1, 1, 1, 1)
+  frames = frames / torch.tensor([0.225, 0.225, 0.255]).view(1, -1, 1, 1, 1)
+  out = model(frames)
+  out = out.mean(0)
+  out1=out.argmax(dim=-1).item()
+  out2=df.iloc[out1,1]
+  return out2
+demo = gr.Interface(fn=recognize, inputs=gr.Video(type="file"),outputs='text',examples= [['dog.mp4']])
 demo.launch()