{"id":29605587,"url":"https://github.com/freedomofkeima/moeflow","last_synced_at":"2025-07-20T16:06:43.219Z","repository":{"id":41678239,"uuid":"110549054","full_name":"freedomofkeima/MoeFlow","owner":"freedomofkeima","description":"Repository for anime characters recognition website, powered by TensorFlow","archived":false,"fork":false,"pushed_at":"2018-09-21T10:13:23.000Z","size":4443,"stargazers_count":119,"open_issues_count":1,"forks_count":7,"subscribers_count":6,"default_branch":"master","last_synced_at":"2023-11-07T20:18:12.542Z","etag":null,"topics":["anime","classification","computer-vision","python3","tensorflow","transfer-learning"],"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/freedomofkeima.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE.md","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null}},"created_at":"2017-11-13T13:03:29.000Z","updated_at":"2023-10-19T07:47:56.000Z","dependencies_parsed_at":"2022-07-09T08:46:11.613Z","dependency_job_id":null,"html_url":"https://github.com/freedomofkeima/MoeFlow","commit_stats":null,"previous_names":[],"tags_count":0,"template":null,"template_full_name":null,"purl":"pkg:github/freedomofkeima/MoeFlow","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/freedomofkeima%2FMoeFlow","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/freedomofkeima%2FMoeFlow/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/freedomofkeima%2FMoeFlow/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/freedomofkeima%2FMoeFlow/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/freedomofkeima","download_url":"https://codeload.github.com/freedomofkeima/MoeFlow/tar.gz/refs/heads/master","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/freedomofkeima%2FMoeFlow/sbom","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":266152700,"owners_count":23884561,"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":["anime","classification","computer-vision","python3","tensorflow","transfer-learning"],"created_at":"2025-07-20T16:06:42.349Z","updated_at":"2025-07-20T16:06:43.210Z","avatar_url":"https://github.com/freedomofkeima.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# MoeFlow\n\n[![CircleCI](https://circleci.com/gh/freedomofkeima/MoeFlow/tree/master.svg?style=shield)](https://circleci.com/gh/freedomofkeima/MoeFlow/tree/master)\n\nRepository for anime characters recognition website, powered by TensorFlow.\n\nDemonstration page (Alpha version): [MoeFlow Website](https://freedomofkeima.com/moeflow/).\n\nMoeFlow is featured in:\n- PyCon ID 2017 | [Presentation](https://freedomofkeima.com/pyconid2017.pdf)\n- [PyCon JP 2018](https://pycon.jp/2018/en/event/sessions) | [Presentation](https://freedomofkeima.com/pyconjp2018.pdf) | [Video](https://www.youtube.com/watch?v=Oh-raRnQoUA)\n\n## Project Introduction\n\nThis project is related to [freedomofkeima/transfer-learning-anime](https://github.com/freedomofkeima/transfer-learning-anime).\n\n### Background\n\nThis project is heavily inspired from characters indexing website such as [saucenao](https://saucenao.com/) and [iqdb](https://www.iqdb.org/). In general, character indexing websites work well since character arts are generally limited in terms of number compared to real-life photos.\n\n\u003cimg alt=\"iqdb_status\" src=\"screenshots/iqdb_status.png\" width=\"600\"\u003e\n\nHowever, there are cases where character indexing websites will not work well, e.g.: the image is cropped or altered.\n\n**Full Image** (Top: Saucenao, Bottom: MoeFlow)\n\n\u003cimg alt=\"saucenao_full_image\" src=\"screenshots/saucenao_from_full_image_2_characters.png\" width=\"600\"\u003e\n\n\u003cimg alt=\"moeflow_full_image\" src=\"screenshots/full_image_2_characters.png\" width=\"600\"\u003e\n\n**Altered Image**\n\n\u003cimg alt=\"saucenao_full_image\" src=\"screenshots/saucenao_from_altered_2_characters.png\" width=\"600\"\u003e\n\n\u003cimg alt=\"moeflow_full_image\" src=\"screenshots/altered_2_characters.png\" width=\"600\"\u003e\n\nOr, there are cases where you want to recognize a character from a photo.\n\n(Top: Saucenao, Middle: iqdb, Bottom: MoeFlow)\n\n\u003cimg alt=\"saucenao_photo\" src=\"screenshots/saucenao_from_photo.png\" width=\"600\"\u003e\n\n\u003cimg alt=\"iqdb_photo\" src=\"screenshots/iqdb_from_photo.png\" width=\"600\"\u003e\n\n\u003cimg alt=\"moeflow_photo\" src=\"screenshots/from_photo.png\" width=\"600\"\u003e\n\n### Transfer Learning\n\nThis project only uses [~~30~~* 60 images per character](100_class_traning_note.md) for learning, which are very low in number for image recognition learning. However, this number is chosen since the majority of characters has a limited number of arts.\n\nIn [yande.re](https://yande.re/tag?name=\u0026order=count\u0026type=4), there are around 35000 registered character tags. However, top 1000 characters only have 70+ images while top 2000 characters only have 40+ images.\n\n(*) The number of dataset is increased from ~30 to ~60 and the overall accuracy is increased by 5% to 10% (from 60% - 65% to 70%).\n\n## Requirements\n\n- TensorFlow 1.4.0 (`pip install tensorflow==1.4.0` first)\n- [nagadomi/animeface-2009](https://github.com/nagadomi/animeface-2009)\n\n## How to create initial environment\n\nPython Environment:\n\n```\n$ virtualenv -p python3 venv  # Ensure python3 version is 3.5, otherwise TensorFlow might not work\n$ . venv/bin/activate\n$ pip install tensorflow==1.4.0\n```\n\nSince `nagadomi/animeface-2009` is an independent project, you need to clone it somewhere in your local directory. Note that the project requires Ruby, ImageMagick, and gcc to run.\n\nAfter you finish installing it, go to `detect.rb` and update the `require` part (line 4) accordingly.\n\nAfter that, you need to download MoeFlow model via `models/download_model.sh` (~ 100 MB).\n\n## How to run\n\nAfter running steps above, you can simply run it by:\n\n```\n$ export MOEFLOW_MODEL_PATH='/path/to/MoeFlow/models'\n$ pip install -e .\n$ app\n```\n\nIf your application is configured to run in a relative path, e.g.: [https://freedomofkeima.com/moeflow/](https://freedomofkeima.com/moeflow/), then you can set static URL path via `export MOEFLOW_RELATIVE_URL_PATH='/moeflow/'`.\n\n## License\n\nThis project itself is licensed under MIT License. \n\nFace recognition feature is developed by [nagadomi](https://github.com/nagadomi).\n\nAll images are owned by their respective creators.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Ffreedomofkeima%2Fmoeflow","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Ffreedomofkeima%2Fmoeflow","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Ffreedomofkeima%2Fmoeflow/lists"}