Alert button

Models Matter, So Does Training: An Empirical Study of CNNs for Optical Flow Estimation

Add code
Bookmark button
Alert button
Sep 14, 2018
Deqing Sun, Xiaodong Yang, Ming-Yu Liu, Jan Kautz

Figure 1 for Models Matter, So Does Training: An Empirical Study of CNNs for Optical Flow Estimation
Figure 2 for Models Matter, So Does Training: An Empirical Study of CNNs for Optical Flow Estimation
Figure 3 for Models Matter, So Does Training: An Empirical Study of CNNs for Optical Flow Estimation
Figure 4 for Models Matter, So Does Training: An Empirical Study of CNNs for Optical Flow Estimation

Share this with someone who'll enjoy it:

Twitter IconFacebook IconLinkedin IconWhatsapp IconMessenger Icon

We investigate two crucial and closely related aspects of CNNs for optical flow estimation: models and training. First, we design a compact but effective CNN model, called PWC-Net, according to simple and well-established principles: pyramidal processing, warping, and cost volume processing. PWC-Net is 17 times smaller in size, 2 times faster in inference, and 11\% more accurate on Sintel final than the recent FlowNet2 model. It is the winning entry in the optical flow competition of the robust vision challenge. Next, we experimentally analyze the sources of our performance gains. In particular, we use the same training procedure of PWC-Net to retrain FlowNetC, a sub-network of FlowNet2. The retrained FlowNetC is 56\% more accurate on Sintel final than the previously trained one and even 5\% more accurate than the FlowNet2 model. We further improve the training procedure and increase the accuracy of PWC-Net on Sintel by 10\% and on KITTI 2012 and 2015 by 20\%. Our newly trained model parameters and training protocols will be available on https://github.com/NVlabs/PWC-Net

View paper onarxiv icon

Share this with someone who'll enjoy it:

Twitter IconFacebook IconLinkedin IconWhatsapp IconMessenger Icon