本文主要是介绍《Image Provenance Analysis at Scale》论文跑通源码,希望对大家解决编程问题提供一定的参考价值,需要的开发者们随着小编来一起学习吧!
1.起因
做毕设正好要做图片溯源(provenance)这方面的内容,因此下载了代码,跑通的过程缺异常痛苦,装包,装虚拟环境,部署代码到服务器等等等等(这些都写过博客),而且跑通代码需要GPU,服务器的GPU被占用就跑不了。而且中间索引步骤每次都要跑30-50分钟,bug多了总是容易放弃,拖拖拉拉花了快一个月才跑完,别的不说,至少看源码能力有了很大的进步。
这里贴上源码的GitHub链接:https://github.com/CVRL/Scalable_Provenance/tree/a5db4f3f51c9cd066e4603959346cae0d86cde18
2.准备工作
首先是最重要的运行程序,runPython3.sh
,这是个shell文件,可以在Linux端用cat查看,下面给出具体内容:
#clean
#rm features/*
#rm index/*
#rm filtering_results/*
#rm filtering_results/json/*#export PYTHONPATH=../notredame/:../featureExtraction/:../indexConstruction/:../provenanceFiltering:../provenanceGraphConstruction:../helperLibraries/:/root/faiss/:/home/jbrogan4/Documents/Projects/Medifor/faiss/#worldIndexFile=/media/jbrogan4/scratch0/medifor/datasets/Nimble/NC2017_Dev1_Beta4/indexes/NC2017_Dev1-provenance-world.csv
#probeIndexFile=/media/jbrogan4/scratch0/medifor/datasets/Nimble/NC2017_Dev1_Beta4/indexes/NC2017_Dev1-provenance-index.csv
#nistDataDir=/media/jbrogan4/scratch0/medifor/datasets/Nimble/NC2017_Dev1_Beta4/
#recall=100#worldIndexFile=data/testset/indexes/test-provenance-world.csv
#probeIndexFile=data/testset/indexes/test-provenance-index.csv
#nistDataDir=data/testsetpython3.7 featureExtractionDriver.py --NISTWorldIndex '/raid/workspace/git_repository/Scalable_Provenance/image_filtering/provenance/tutorial/data/testset/indexes/test-provenance-world.csv' --NISTDataset '/raid/workspace/git_repository/Scalable_Provenance/image_filtering/provenance/tutorial/data/testset' --outputdir features/#mkdir tmp
ls -d features/* > /raid/workspace/git_repository/Scalable_Provenance/image_filtering/provenance/tutorial/tmp/feature_filelist#mkdir indexTraining
这篇关于《Image Provenance Analysis at Scale》论文跑通源码的文章就介绍到这儿,希望我们推荐的文章对编程师们有所帮助!