cv2.resize slows down the training time largely

[GunjanDhanuka]

First of all, thanks a lot for sharing your code! I noticed that in the training pipeline (video_dataset.py), you have used the following:

width, height = self.quantize_size, channels
features = cv2.resize(features, (width, height),
    interpolation=cv2.INTER_LINEAR) # CxTxN

This resizing delays the training pipeline by a lot. Instead of resizing on every 32 frames, shouldn't we just do it on the last some frames which are less than 32?