Comb, Prune, Distill: Towards Unified Pruning for Vision Model Compression
This addresses the need for efficient models on resource-limited devices like intelligent vehicles, offering a generalizable solution across architectures and tasks, though it is incremental in improving existing pruning methods.
The paper tackles the problem of model compression for vision tasks by proposing a unified pruning framework called Comb, Prune, Distill (CPD), which achieves a speedup of up to x4.3 with 1.8% accuracy loss in image classification and up to x1.89 with 5.1% mIoU loss in semantic segmentation.
Lightweight and effective models are essential for devices with limited resources, such as intelligent vehicles. Structured pruning offers a promising approach to model compression and efficiency enhancement. However, existing methods often tie pruning techniques to specific model architectures or vision tasks. To address this limitation, we propose a novel unified pruning framework Comb, Prune, Distill (CPD), which addresses both model-agnostic and task-agnostic concerns simultaneously. Our framework employs a combing step to resolve hierarchical layer-wise dependency issues, enabling architecture independence. Additionally, the pruning pipeline adaptively remove parameters based on the importance scoring metrics regardless of vision tasks. To support the model in retaining its learned information, we introduce knowledge distillation during the pruning step. Extensive experiments demonstrate the generalizability of our framework, encompassing both convolutional neural network (CNN) and transformer models, as well as image classification and segmentation tasks. In image classification we achieve a speedup of up to x4.3 with a accuracy loss of 1.8% and in semantic segmentation up to x1.89 with a 5.1% loss in mIoU.