{"id":17267159,"url":"https://github.com/agamiko/neural-based-data-augmentation","last_synced_at":"2025-03-26T11:13:41.231Z","repository":{"id":101298801,"uuid":"201916892","full_name":"AgaMiko/neural-based-data-augmentation","owner":"AgaMiko","description":"Improving generalization via style transfer-based data augmentation: Novel regularization method","archived":false,"fork":false,"pushed_at":"2019-09-02T09:59:28.000Z","size":3266,"stargazers_count":1,"open_issues_count":0,"forks_count":0,"subscribers_count":3,"default_branch":"master","last_synced_at":"2025-01-31T12:22:19.425Z","etag":null,"topics":["cnn","data-augmentation","database","deep-learning","deep-neural-networks","image-classification","image-synthesis","paper","skin-lesions","style-transfer"],"latest_commit_sha":null,"homepage":null,"language":null,"has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/AgaMiko.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"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":"2019-08-12T11:26:29.000Z","updated_at":"2020-04-05T14:17:20.000Z","dependencies_parsed_at":null,"dependency_job_id":"eb0fffad-c367-45a1-a2dc-ec568a7d5308","html_url":"https://github.com/AgaMiko/neural-based-data-augmentation","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/AgaMiko%2Fneural-based-data-augmentation","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/AgaMiko%2Fneural-based-data-augmentation/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/AgaMiko%2Fneural-based-data-augmentation/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/AgaMiko%2Fneural-based-data-augmentation/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/AgaMiko","download_url":"https://codeload.github.com/AgaMiko/neural-based-data-augmentation/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":245641437,"owners_count":20648644,"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","data-augmentation","database","deep-learning","deep-neural-networks","image-classification","image-synthesis","paper","skin-lesions","style-transfer"],"created_at":"2024-10-15T08:09:36.043Z","updated_at":"2025-03-26T11:13:41.212Z","avatar_url":"https://github.com/AgaMiko.png","language":null,"funding_links":[],"categories":[],"sub_categories":[],"readme":"\n# Improving generalization via style transfer-based data augmentation: Novel regularization method\n\n![Generated skin lesions: an example](https://github.com/AgaMiko/ST-DA/blob/master/Skin-lesions-examples.jpg)\n\n## Introduction\nCurrently, deep learning  algorithms are considered as state-of-the-art in many classification tasks,\nand yet the problem of weak generalization is very common, widely mentioned, and still up-to-date.\n\nThe present paper focuses most on the data augmentation. In our method, new images are synthetized with \u003cem\u003eneural style transfer (NST)\u003c/em\u003e,\nand the generated images are then used to train the convolutional neural network (CNN) in order to improve\nits generalization abilities.  \nThe main contributions of this paper are:\n*\tThe proposition of using \u003cem\u003eneural style transfer\u003c/em\u003e for the data augmentation (ST-DA). This approach is presented on the skin lesion case study by transforming a benign skin lesion to a malignant lesion, and tested with dataset enrichment evaluation; \n*\tIncorporating unlabeled, synthesized data into training by adding \u003cem\u003epseudo-labels\u003c/em\u003e generated by another CNN; \n*\tLimiting the problem of noisy \u003cem\u003epseudo-labels\u003c/em\u003e in synthetic images used as a CNN training set by using only real images in validation and test sets;\n*\tEvaluating the ability to enrich the training dataset with artificially generated data with \u003cem\u003eDeep Taylor Decomposition\u003c/em\u003e, \n* Proving that the ST-DA method significantly improves the performance and repeatability of training for deep neural networks.\n\n\n## ST-DA\n### How-to\nShort and friendly how-to tutorial will be soon available [here](https://github.com/AgaMiko/ST-DA/blob/master/images/instruction.md)\n\n### Details\nThe result and details of the method will be able to be find soon in the original paper here: [soon](xxx)\nYou can check instead our previous papers about data augmentation:\n  * [Data augmentation for improving deep learning in image classification problem, 2018](https://ieeexplore.ieee.org/abstract/document/8388338)\n  * [Style transfer-based image synthesis as an efficient regularization technique in deep learning, 2019](https://arxiv.org/abs/1905.10974)\n\n## Database \n### Download\nThe total databse size is 248 489 unalabeled generated dermoscopic images of skin lesions (224x224 px). \n* Few full-size examples can be found [here](https://github.com/AgaMiko/ST-DA/tree/master/images)\n* Database can be download [soon here](xxx) (soon)\n#### If you use this database please star the repository and cite the following paper (soon):\n\u003cem\u003e [\"Improving generalization via style transfer-based data augmentation: Novel regularization method\"](xxxx)\u003c/em\u003e, by [Agnieszka Mikołajczyk](https://scholar.google.pl/citations?user=VFMjpTsAAAAJ\u0026hl=en) , [Michał Grochowski](https://scholar.google.pl/citations?user=UTA55L8AAAAJ\u0026hl=en), [Arkadiusz Kwasigroch](https://scholar.google.pl/citations?user=Hw7DV4QAAAAJ\u0026hl=en)\n\n## Sources\n\nThe database was generated using following sources:\n\n* *Image generation:*\n  * **Style transfer original paper:** [A Neural Algorithm of Artistic Style](https://arxiv.org/abs/1508.06576) is a first paper that presented \u003cem\u003eNeural Style Transfer\u003c/em\u003e. \n  * **Style transfer implementation:** Implementation of [Neural Style Transfer \u0026 Neural Doodles](https://github.com/titu1994/Neural-Style-Transfer) from the paper \u003cem\u003eA Neural Algorithm of Artistic Style\u003c/em\u003e in Keras 2.0+\n* *Explainability method:*\n  * **Deep Taylor decomposition:** [DeepTaylor](https://www.sciencedirect.com/science/article/pii/S0031320316303582?via%3Dihub) computes for each neuron a rootpoint, that is close to the input, but which's output value is 0, and uses this difference to estimate the attribution of each neuron recursively.\n   * **Repository:** [iNNvestigate](https://github.com/albermax/innvestigate) library contains implementations for the\n   SmoothGrad, DeConvNet, Guided BackProp,  PatternNet, DeepTaylor, PatternAttribution, IntegratedGradients and DeepLIFT.  \n* *Source database:*\n  * **ISIC Archive:** The [ISIC Archive](https://www.isic-archive.com) contains over 23k images of skin lesions, labeled as 'benign' or 'malignant'. Those images were used to generate our database.\n  * **ISIC Archive Downloader:** A [script](https://github.com/GalAvineri/ISIC-Archive-Downloader) to download the ISIC Archive of lesion images \n* *Previous papers about data augmentation:*\n  * [Data augmentation for improving deep learning in image classification problem, 2018](https://ieeexplore.ieee.org/abstract/document/8388338)\n  * [Style transfer-based image synthesis as an efficient regularization technique in deep learning, 2019](https://arxiv.org/abs/1905.10974)\n* *Similar projects:*\n  * **Generating skin lesions with GANs** - [Beating Melanoma with Deep Learning: letting the data speak](https://github.com/devansh20la/Beating-Melanoma/tree/master/Generator)\n* *Other:*\n  * **VGG8** [Selected Technical Issues of Deep Neural Networks for Image Classification Purposes](http://www.czasopisma.pan.pl/Content/112085/PDF/21_363-376_00946_Bpast.No.67-2_28.04.19_K3.pdf) prestents the details of VGG8 architecture.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fagamiko%2Fneural-based-data-augmentation","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fagamiko%2Fneural-based-data-augmentation","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fagamiko%2Fneural-based-data-augmentation/lists"}