{"id":16919551,"url":"https://github.com/bobholamovic/cnn-friqa","last_synced_at":"2025-04-11T16:38:54.592Z","repository":{"id":122357806,"uuid":"160135698","full_name":"Bobholamovic/CNN-FRIQA","owner":"Bobholamovic","description":"Convolutional Neural Network for Full-Reference color Image Quality Assessment","archived":false,"fork":false,"pushed_at":"2019-06-03T04:29:12.000Z","size":63,"stargazers_count":16,"open_issues_count":0,"forks_count":6,"subscribers_count":1,"default_branch":"master","last_synced_at":"2025-03-25T12:50:52.612Z","etag":null,"topics":["cnn","iqa","pytorch"],"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/Bobholamovic.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":"2018-12-03T05:09:51.000Z","updated_at":"2024-03-31T08:55:14.000Z","dependencies_parsed_at":null,"dependency_job_id":"339d4daf-417a-4c94-b6c8-2e820c8fa420","html_url":"https://github.com/Bobholamovic/CNN-FRIQA","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/Bobholamovic%2FCNN-FRIQA","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Bobholamovic%2FCNN-FRIQA/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Bobholamovic%2FCNN-FRIQA/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Bobholamovic%2FCNN-FRIQA/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/Bobholamovic","download_url":"https://codeload.github.com/Bobholamovic/CNN-FRIQA/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":248441884,"owners_count":21104094,"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":["cnn","iqa","pytorch"],"created_at":"2024-10-13T19:44:44.409Z","updated_at":"2025-04-11T16:38:54.566Z","avatar_url":"https://github.com/Bobholamovic.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# CNN-FRIQA\nConvolutional Neural Network for Full-Reference color Image Quality Assessment  \n\nThis is a lightweight version of [deepIQA](https://arxiv.org/abs/1612.01697). The original project of deepIQA is [here](https://github.com/dmaniry/deepIQA) and the paper is [here](https://arxiv.org/abs/1612.01697). \n  \n  \n## Environment and Dependencies\n\u003e Ubuntu 16.04 64-bit, Visual Studio Code, Python 3.5.2, Pytorch 0.4.0\n\n`requirements.txt` is not included yet.   \n  \n  \n  \n## Usage\n\n### Data Preparation\nNow the data lists are stored in `.json` files. The relative paths of the distorted images and the reference images to `data-dir` and the quality scores (ground-truth values) are contained in three arrays of a `json` object, with the fields specified as `img`, `ref`, and `score`, respectively. For example, `train_data.json` may look like this:\n\n```\n{\n  \"img\":\n    [\n      \"distorted/img11_2_4.bmp\", \n      \"distorted/img6_3_3.bmp\"\n    ], \n  \"ref\":\n    [\n      \"images/img11.bmp\", \n      \"distorted/img6.bmp\"\n    ], \n  \"score\":\n    [\n      0.5503, \n      0.4312\n    ]\n}\n```\n(this has been prettified as everthing actually on one line)\n\nAlso, there are `val_data.json` for validation subset and `test_data.json` for test subset. The lists are expected to be found at `list-dir`, which will be set to `data-dir` if not specified. \n\nThe scripts for data preparation on `Waterloo` and `TID2013` are provided. \n\n### Running Code\nStart from the root directory of this project, \n```bash\ncd src/\n```\n\nFor training, try\n```bash\npython iqa.py train --resume pretrianed_model_path --data-dir DIR_OF_DATASET\n```\n\nIf `pretrained_model_path` is not correctly specified, the model will learn from scratch. \n\nUse\n```bash\npython iqa.py train --resume pretrianed_model_path | tee train.log\n```\nto dump logs. \n  \nFor evaluation, try\n```bash\npython iqa.py train --evaluate --resume pretrained_model_path\n```\n  \nFor testing, try\n```bash\npython iqa.py test --resume pretrained_model_path\n```\n  \nThe code of testing the model on a single image is desired, yet to be provided. \n\nAs the patches are randomly extracted, there should be a random noise in the output of the model, which explains the slight difference of the performances upon different attempts. \n\nSome pertrained models ~~and the script to make filename lists~~ are to be uploaded later.   \n  \n  \n  \n## Experiment and Performance\nRoughly, the `SROCC` value reaches `0.95` or higher under the best condition.\n\nThe experiment results are to be added here.  \n  \n  \n  \n## Acknowledgement\n+ The design of the model is based on [Deep Neural Networks for No-Reference and Full-Reference Image Quality Assessment](https://arxiv.org/abs/1612.01697)\n+ Torch version of `MS-SSIM` from [lizhengwei1992/MS_SSIM_pytorch](https://github.com/lizhengwei1992/MS_SSIM_pytorch.git)\n+ Part of the code layout from [fyu/drn](https://github.com/fyu/drn)\n\nWith best thanks!  \n\n  \n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fbobholamovic%2Fcnn-friqa","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fbobholamovic%2Fcnn-friqa","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fbobholamovic%2Fcnn-friqa/lists"}