Memory-Efficient Implementation of DenseNets
This work addresses memory bottlenecks for researchers and practitioners training deep neural networks, particularly DenseNets, allowing for more efficient and scalable model development.
The authors tackled the high GPU memory consumption of DenseNets during training by introducing strategies that reduce memory cost from quadratic to linear, enabling training of deeper networks such as a 264-layer DenseNet with 73M parameters on a single workstation and achieving a state-of-the-art top-1 error of 20.26% on ImageNet.
The DenseNet architecture is highly computationally efficient as a result of feature reuse. However, a naive DenseNet implementation can require a significant amount of GPU memory: If not properly managed, pre-activation batch normalization and contiguous convolution operations can produce feature maps that grow quadratically with network depth. In this technical report, we introduce strategies to reduce the memory consumption of DenseNets during training. By strategically using shared memory allocations, we reduce the memory cost for storing feature maps from quadratic to linear. Without the GPU memory bottleneck, it is now possible to train extremely deep DenseNets. Networks with 14M parameters can be trained on a single GPU, up from 4M. A 264-layer DenseNet (73M parameters), which previously would have been infeasible to train, can now be trained on a single workstation with 8 NVIDIA Tesla M40 GPUs. On the ImageNet ILSVRC classification dataset, this large DenseNet obtains a state-of-the-art single-crop top-1 error of 20.26%.