Co-Training Transformer with Videos and Images Improves Action Recognitionai.googleblog.com2 pointstokyopanda4 years ago