FILM

FILM: Frame Interpolation for Large Scene Motion Overview

FILM, or Frame Interpolation for Large Scene Motion, is a high-quality frame interpolation neural network implemented using Tensorflow 2. This innovative tool presents a unified single-network approach that does not require the use of additional pre-trained networks, such as optical flow or depth. Despite this, it still manages to achieve state-of-the-art results. FILM utilizes a multi-scale feature extractor that shares the same convolution weights across the scales and is trainable from frame triplets alone. This tool was developed by a team of researchers from Google and was detailed in a technical report in 2022.

FILM Highlights

• It presents a unified single-network approach that doesn't require additional pre-trained networks.
• It uses a multi-scale feature extractor that shares the same convolution weights across the scales.
• The model is trainable from frame triplets alone, making it highly efficient and versatile.