CondConv: Conditionally Parameterized Convolutions for Efficient Inference
This work addresses the efficiency and accuracy trade-off in deep learning models for tasks like image classification and detection, offering a novel method to scale networks without increasing inference costs, though it is incremental in the context of convolutional layer improvements.
The paper tackles the problem of improving the performance and inference cost trade-off in convolutional neural networks by proposing conditionally parameterized convolutions (CondConv), which learn specialized kernels for each example instead of sharing them across all examples. This approach applied to EfficientNet-B0 achieves state-of-the-art performance of 78.3% accuracy on ImageNet classification with only 413M multiply-adds.
Convolutional layers are one of the basic building blocks of modern deep neural networks. One fundamental assumption is that convolutional kernels should be shared for all examples in a dataset. We propose conditionally parameterized convolutions (CondConv), which learn specialized convolutional kernels for each example. Replacing normal convolutions with CondConv enables us to increase the size and capacity of a network, while maintaining efficient inference. We demonstrate that scaling networks with CondConv improves the performance and inference cost trade-off of several existing convolutional neural network architectures on both classification and detection tasks. On ImageNet classification, our CondConv approach applied to EfficientNet-B0 achieves state-of-the-art performance of 78.3% accuracy with only 413M multiply-adds. Code and checkpoints for the CondConv Tensorflow layer and CondConv-EfficientNet models are available at: https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet/condconv.