Introducing FILM: Frame Interpolation for Large Scene Motion. This Tensorflow 2 implementation neural network achieves state-of-the-art results without using additional pre-trained networks. The model is trainable from frame triplets alone, and predictions run on Nvidia T4 GPU hardware. With a predict time of typically 11 seconds, this model is perfect for large scene motion. Get started now and explore the paper, benchmark scores, and view more examples.
Share: