{"id":23065559,"url":"https://github.com/mtli/llcv","last_synced_at":"2025-08-23T22:32:10.794Z","repository":{"id":57440575,"uuid":"349317344","full_name":"mtli/llcv","owner":"mtli","description":"An Extensible Framework for Low-Latency Computer Vision Research :fast_forward:","archived":false,"fork":false,"pushed_at":"2022-04-26T16:02:03.000Z","size":102,"stargazers_count":6,"open_issues_count":0,"forks_count":0,"subscribers_count":1,"default_branch":"main","last_synced_at":"2024-12-15T23:20:33.108Z","etag":null,"topics":["classification","computer-vision","deep-learning","detection","latency"],"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/mtli.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}},"created_at":"2021-03-19T05:59:20.000Z","updated_at":"2024-03-05T18:00:50.000Z","dependencies_parsed_at":"2022-09-26T17:20:42.718Z","dependency_job_id":null,"html_url":"https://github.com/mtli/llcv","commit_stats":null,"previous_names":["mtli/mecv"],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mtli%2Fllcv","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mtli%2Fllcv/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mtli%2Fllcv/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/mtli%2Fllcv/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/mtli","download_url":"https://codeload.github.com/mtli/llcv/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":230745898,"owners_count":18274171,"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":["classification","computer-vision","deep-learning","detection","latency"],"created_at":"2024-12-16T05:08:57.693Z","updated_at":"2024-12-21T17:23:58.396Z","avatar_url":"https://github.com/mtli.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# LLCV \u0026mdash; An Extensible Framework for **L**ow-**L**atency **C**omputer **V**ision Research\r\n\r\n\r\n![#fc4903](https://via.placeholder.com/15/fc4903/000000?text=+) Research-oriented \u0026mdash; flexible interfaces and dynamic modules\r\n\r\n![#3865cf](https://via.placeholder.com/15/3865cf/000000?text=+) Latency-oriented \u0026mdash; efficient implementation and various types of timing support\r\n\r\nFeatures:\r\n- Training auto-resume after interruption\r\n- Support for DataParellel \u0026 DistributedDataParallel\r\n- Tensorboard integration\r\n\r\nSupported tasks:\r\n- Image classification\r\n- Object detection\r\n\r\nNotable models:\r\n- ResNet-18 on CIFAR-10 (Top-1 95.62%)\r\n- ResNet-50 on ImageNet (Top-1 76.56%, Top-5 93.02%)\r\n- Faster R-CNN on COCO (AP 37.0, through torchvision)\r\n\r\nNote that this repo is currently under active development and compatibility is not guaranteed. While there are many standard techniques for faster implementations, this repo mostly focuses on the research aspect of reducing training and inference cost.\r\n\r\n\r\n## Installation\r\n\r\n[![PyPI version](https://badge.fury.io/py/llcv.svg)](https://badge.fury.io/py/llcv)\r\n\r\nInstall [PyTorch](https://pytorch.org/) (\u003e= 1.1) and run:\r\n```\r\npip install llcv\r\n```\r\n\r\nIf you prefer to install it in develop mode:\r\n```\r\ngit clone https://github.com/mtli/llcv\r\ncd llcv\r\npip install -e .\r\n```\r\n\r\n\r\n## Getting Started\r\n\r\nCheck out the scripts in the `samples/` for various tasks.\r\n\r\nTo use this framework for your own projects, you can either fork this repo, or using this repo as a dependency through [dynamic modules](doc/design.md#dynamic-modules).\r\n\r\nYou can find more details about the framework's design [here](doc/design.md).\r\n\r\n\r\n## Model Zoo\r\n\r\nNote that latency and throughput are measured on Geforce GTX 1080 Ti GPUs, unless otherwise stated.\r\n\r\n\u003cbr\u003e\r\n\r\n### CIFAR-10 (Classification)\r\n| ID  |    Method    | Epc | Top-1 (%)  | E2E (ms) | Inf (ms) | Train BS | Train TP | Train Time | Ckpt | Log | Script |\r\n| :-: | :----------: | :-: | :----: | :------: | :------: | :------: | :------: | :--------: | :--: | :-: | :----: |\r\n| 1   | ResNet-18    | 200 | 95.62 |        - | 3.05 ± 0.302 |  128 |   2081Hz |      1h22m | [ckpt](https://www.cs.cmu.edu/~mengtial/proj/llcv/model_zoo/c10-r18-e200-95.62-870a16f.pth) | [log](https://www.cs.cmu.edu/~mengtial/proj/llcv/model_zoo/c10-r18-e200-95.62-870a16f.log) | [script](samples/cifar10/c10.sh) |\r\n\r\nIn the above table, timings are measured on a single GPU for both training and testing.\r\n\r\n\u003cbr\u003e\r\n\r\n### ImageNet (Classification)\r\n| ID  |    Method    | Epc | Top-1 (%) | Top-5 (%) | E2E (ms) |  Inf (ms) | Train BS | Train TP | Train Time | Ckpt | Log | Script |\r\n| :-: | :----------: | :-: | :-------: | :-------: |:-------: | :-------: | :------: | :------: | :--------: | :--: | :-: | :----: |\r\n| 1   | ResNet-50    | 90  |     76.56 |     93.02 |        - | 9.82 ± 3.59 |    256 |  504.8Hz |      2d16h | [ckpt](https://www.cs.cmu.edu/~mengtial/proj/llcv/model_zoo/im-r50-e90-76.56-870a16f.pth) | [log](https://www.cs.cmu.edu/~mengtial/proj/llcv/model_zoo/im-r50-e90-76.56-870a16f.log) | [script](samples/imagenet/im_r50.sh) |\r\n\r\nIn the above table, timings are measured on 4 GPUs for training and a single GPU for testing.\r\n\r\n\u003cbr\u003e\r\n\r\n### COCO (Detection)\r\n\r\n| ID  |    Method    | Epc |  AP  | E2E (ms) |  Inf (ms) | Train BS | Train TP | Train Time | Ckpt | Log | Script |\r\n| :-: | :----------: | :-: | :--: |:-------: | :-------: | :------: | :------: | :--------: | :--: | :-: | :----: |\r\n| 1   | Faster R-CNN R50 FPN (torchvision)  | - | 37.0 | - | 75.2 ± 54.4 | - | - |        - | -    | -   | [script](samples/coco/coco_frcnn_test_pretrained.sh) |\r\n\r\nIn the above table, timings are measured on 8 GPUs for training and a single GPU for testing.\r\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmtli%2Fllcv","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fmtli%2Fllcv","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmtli%2Fllcv/lists"}