Improving the Resolution of CNN Feature Maps Efficiently with Multisampling
This addresses a bottleneck in neural network performance for image classification, offering an efficient solution with broad applicability.
The paper tackles the problem of information loss in CNN feature maps during subsampling by introducing multisampling, which improves accuracy in models like DenseNet and ResNet without extra parameters, and even boosts pretrained ImageNet models without training.
We describe a new class of subsampling techniques for CNNs, termed multisampling, that significantly increases the amount of information kept by feature maps through subsampling layers. One version of our method, which we call checkered subsampling, significantly improves the accuracy of state-of-the-art architectures such as DenseNet and ResNet without any additional parameters and, remarkably, improves the accuracy of certain pretrained ImageNet models without any training or fine-tuning. We glean possible insight into the nature of data augmentations and demonstrate experimentally that coarse feature maps are bottlenecking the performance of neural networks in image classification.