Dr$^2$Net: Dynamic Reversible Dual-Residual Networks for Memory-Efficient Finetuning
This addresses memory constraints for researchers and practitioners in computer vision working with high-resolution data like video or point clouds, though it is incremental as it builds on reversible network concepts.
The paper tackled the problem of memory-intensive finetuning of large pretrained models for high-resolution tasks by proposing Dr$^2$Net, a reversible network architecture that reduces memory consumption while achieving comparable performance, with significantly less memory usage reported.
Large pretrained models are increasingly crucial in modern computer vision tasks. These models are typically used in downstream tasks by end-to-end finetuning, which is highly memory-intensive for tasks with high-resolution data, e.g., video understanding, small object detection, and point cloud analysis. In this paper, we propose Dynamic Reversible Dual-Residual Networks, or Dr$^2$Net, a novel family of network architectures that acts as a surrogate network to finetune a pretrained model with substantially reduced memory consumption. Dr$^2$Net contains two types of residual connections, one maintaining the residual structure in the pretrained models, and the other making the network reversible. Due to its reversibility, intermediate activations, which can be reconstructed from output, are cleared from memory during training. We use two coefficients on either type of residual connections respectively, and introduce a dynamic training strategy that seamlessly transitions the pretrained model to a reversible network with much higher numerical precision. We evaluate Dr$^2$Net on various pretrained models and various tasks, and show that it can reach comparable performance to conventional finetuning but with significantly less memory usage.