Performance Analysis of Digital Pathology Pipeline (ETA 8/27) #2728
Description
Is your feature request related to a problem? Please describe.
Analyzing the performance of digital pathology pipeline in order to measure the computation time for each involving components and identify the bottlenecks. This information should be used as the basis to improve the performance of this pipeline.
Describe the solution you'd like
Developing the tooling/guides and end-to-end performance analysis of the digital pathology pipeline. This involves first adding timestamp markers for start/end of important events/components, like data loading, pre/post transformations, and the neural network model to existing digital pathology pipelines. Then profiling tools such as DLProf and Nsight Systems can be utilized to measure CPU/GPU computation time for each of these component and overall GPU utilization and memory consumption to identify the bottleneck and rooms for improvement.