# Modify the model to output features num_ftrs = model.fc.in_features model.fc = nn.Identity() # Replace the classification layer with an identity function
print(features.shape) The extracted features can be used for various downstream tasks such as video clustering, similarity search, classification, etc. anal friend request.mp4
# Extract features with torch.no_grad(): features = model(video_tensor) # Modify the model to output features num_ftrs = model