Mix & Match: training convnets with mixed image sizes for improved accuracy, speed and scale resiliency
This work addresses the need for more efficient and scalable CNN training and inference in computer vision, offering a method to accelerate training and improve accuracy with mixed image sizes, though it is incremental as it builds on existing CNN architectures and training practices.
The paper tackles the problem of training convolutional neural networks with a fixed image size by introducing a mixed-size training regime that improves model resilience to image size changes and generalization on small images, achieving a 76.43% top-1 accuracy with ResNet50 at image size 160 (matching baseline accuracy with 2x fewer computations) and a 79.27% accuracy at size 288 (14% relative improvement over baseline).
Convolutional neural networks (CNNs) are commonly trained using a fixed spatial image size predetermined for a given model. Although trained on images of aspecific size, it is well established that CNNs can be used to evaluate a wide range of image sizes at test time, by adjusting the size of intermediate feature maps. In this work, we describe and evaluate a novel mixed-size training regime that mixes several image sizes at training time. We demonstrate that models trained using our method are more resilient to image size changes and generalize well even on small images. This allows faster inference by using smaller images attest time. For instance, we receive a 76.43% top-1 accuracy using ResNet50 with an image size of 160, which matches the accuracy of the baseline model with 2x fewer computations. Furthermore, for a given image size used at test time, we show this method can be exploited either to accelerate training or the final test accuracy. For example, we are able to reach a 79.27% accuracy with a model evaluated at a 288 spatial size for a relative improvement of 14% over the baseline.