On Vectorization of Deep Convolutional Neural Networks for Vision Tasks
Vectorization (mathematics)
Implementation
Speedup
DOI:
10.1609/aaai.v29i1.9488
Publication Date:
2022-06-23T19:04:28Z
AUTHORS (2)
ABSTRACT
We recently have witnessed many ground-breaking results in machine learning and computer vision, generated by using deep convolutional neural networks (CNN). While the success mainly stems from large volume of training data network architectures, vector processing hardware (e.g. GPU) undisputedly plays a vital role modern CNN implementations to support massive computation. Though much attention was paid extent literature understand algorithmic side CNN, little research dedicated vectorization for scaling up CNNs. In this paper, we studied process key building blocks CNNs, order better facilitate parallel implementation. Key steps testing CNNs are abstracted as matrix operators, upon which parallelism can be easily achieved. developed compared six with various degrees illustrated impact on speed model testing. Besides, unified framework both high-level low-level vision tasks is provided, along vectorized Matlab implementation state-of-the-art performance.
SUPPLEMENTAL MATERIAL
Coming soon ....
REFERENCES (0)
CITATIONS (33)
EXTERNAL LINKS
PlumX Metrics
RECOMMENDATIONS
FAIR ASSESSMENT
Coming soon ....
JUPYTER LAB
Coming soon ....