{"id":22354765,"url":"https://github.com/pprp/mixed_precision_imagenet_benchmark","last_synced_at":"2025-03-26T12:41:55.774Z","repository":{"id":112699091,"uuid":"316745285","full_name":"pprp/mixed_precision_imagenet_benchmark","owner":"pprp","description":"ResNet50在ImageNet上混合精度","archived":false,"fork":false,"pushed_at":"2021-06-10T12:27:35.000Z","size":3302,"stargazers_count":4,"open_issues_count":0,"forks_count":0,"subscribers_count":2,"default_branch":"main","last_synced_at":"2025-01-31T13:43:57.598Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/pprp.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2020-11-28T13:56:17.000Z","updated_at":"2023-08-15T08:46:32.000Z","dependencies_parsed_at":null,"dependency_job_id":"553c0d4c-e286-4777-b666-bc0b29d7663d","html_url":"https://github.com/pprp/mixed_precision_imagenet_benchmark","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/pprp%2Fmixed_precision_imagenet_benchmark","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/pprp%2Fmixed_precision_imagenet_benchmark/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/pprp%2Fmixed_precision_imagenet_benchmark/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/pprp%2Fmixed_precision_imagenet_benchmark/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/pprp","download_url":"https://codeload.github.com/pprp/mixed_precision_imagenet_benchmark/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":245658981,"owners_count":20651519,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2024-12-04T13:15:07.610Z","updated_at":"2025-03-26T12:41:55.768Z","avatar_url":"https://github.com/pprp.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# mixed_precision_pytorch\n\nauthor: pprp\n\ndate: 2021-1-1\n\n## 项目目的\n\n本项目主要是为了复现ResNet50在ImageNet上的结果，并且测试混合精度训练对模型的影响，分析其对训练速度、训练准确率等指标的影响。\n\n## 项目介绍\n\n数据来源：\n\nhttp://www.image-net.org/download-images\n\n主要有两个版本：\n\n- project_lightning: 使用了Pytorch Lightning框架，实现了训练ImageNet。\n- project_pytorch: 使用了原生PyTorch配合apex进行ImageNet的训练。\n- requirements.txt列举了环境的依赖。\n- run_official.sh是训练时运行的脚本。\n\n数据集、log文件都没有包括在库中。\n\n硬件：\n\n- CPU：Intel Core I9-10900K @ 3.70GHz \n\n- 显卡：GeForce RTX 2080Ti(11G) x 4 \n\n- 内存：64G \n\n\n软件： \n\n- 系统：Window 10 \n- CUDA 11.0+CuDNN 8.0 \n- PyTorch 1.6.0、Apex、PyTorch Lightning \n- 容器：Anaconda \n- IDE：Vscode\n\n## 实验结果\n\n训练过程top1 acc: \n\n![](images/train_acc1.png)\n\n训练过程top5 acc: \n\n![](images/train_acc5.png)\n\n训练过程中loss:\n\n![](images/train_loss.png)\n\n验证集上top1 acc:\n\n![](images/val_acc1.png)\n\n验证集上top5 acc:\n\n![](images/val_acc5.png)\n\n验证集loss曲线：\n\n![](images/val_loss.png)\n\n\n\n| 指标                 | 混合精度 | FP32训练 |\n| -------------------- | -------- | -------- |\n| Top1 Accuracy(train) | 79.19%   | 78.94%   |\n| Top2 Accuracy(train) | 92.78%   | 92.62%   |\n| Top1 Accuracy(val)   | 75.63%   | 75.46%   |\n| Top2 Accuracy(val)   | 92.68%   | 92.61%   |\n| Time-consuming       | 2day     | 2day12h  |\n\n## 参考文献\n\n[1] He, Kaiming, Zhang, Xiangyu, Ren, Shaoqing and Sun, Jian Deep Residual Learning for Image Recognition. (2015). , cite arxiv:1512.03385Comment: Tech report .\n\n[2] Micikevicius, P.; Narang, S.; Alben, J.; Diamos, G.; Elsen, E.; Garcia, D.; Ginsburg, B.; Houston, M.; Kuchaiev, O.; Venkatesh, G. \u0026 Wu, H. (2017), 'Mixed Precision Training' , cite arxiv:1710.03740Comment: Published as a conference paper at ICLR 2018 .\n\n[3] He, T.; Zhang, Z.; Zhang, H.; Zhang, Z.; Xie, J. \u0026 Li, M. (2018), 'Bag of Tricks for Image Classification with Convolutional Neural\n Networks', .\n\n ","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fpprp%2Fmixed_precision_imagenet_benchmark","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fpprp%2Fmixed_precision_imagenet_benchmark","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fpprp%2Fmixed_precision_imagenet_benchmark/lists"}