Computer Vision News - February 2020

Summary We Tried for You 16 Here we can see that Efficient-Net-B7 is much more accurate, however the running time has increased. This is obviously due to the B7 model being much larger. Note that the running time of Efficient-Net can vary between different environments. Moreover, the general scheme of the model can be used to train your own model with the constraints of your resources. On the other hand, this small experiment demonstrates that FLOPs are not always a guarantee for running time. The architecture and implementation of the model can alter the running time too. Efficient-Net is a novel framework for scaling your models. It suggests an efficient way to choose the architecture that best fits your needs. In this article we demonstrated how to run Efficient-Net with PyTorch and we performed a small experiment to compare it to its baseline models - ResNet and MobileNet. We saw that, although theoretically it performs fewer FLOPs than the baseline model, it does not always run faster. This, however, can be improved by adjusting the model to your own device. "Although theoretically it performs fewer FLOPs than the baseline model, it does not always run faster." Conclusion