{"id":32387933,"url":"https://github.com/shakshi3104/tfmars","last_synced_at":"2026-05-19T05:31:33.096Z","repository":{"id":43065402,"uuid":"448465787","full_name":"Shakshi3104/tfmars","owner":"Shakshi3104","description":"MarNASNets and CNN for sensor-based human activity recognition built in TensorFlow","archived":false,"fork":false,"pushed_at":"2023-05-12T10:11:44.000Z","size":16047,"stargazers_count":2,"open_issues_count":0,"forks_count":0,"subscribers_count":3,"default_branch":"master","last_synced_at":"2023-05-12T11:25:50.826Z","etag":null,"topics":["convolutional-neural-networks","human-activity-recognition","tensorflow"],"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/Shakshi3104.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":"2022-01-16T05:27:55.000Z","updated_at":"2023-05-12T11:25:50.827Z","dependencies_parsed_at":"2023-01-24T03:15:18.600Z","dependency_job_id":null,"html_url":"https://github.com/Shakshi3104/tfmars","commit_stats":null,"previous_names":[],"tags_count":null,"template":null,"template_full_name":null,"purl":"pkg:github/Shakshi3104/tfmars","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Shakshi3104%2Ftfmars","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Shakshi3104%2Ftfmars/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Shakshi3104%2Ftfmars/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Shakshi3104%2Ftfmars/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/Shakshi3104","download_url":"https://codeload.github.com/Shakshi3104/tfmars/tar.gz/refs/heads/master","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Shakshi3104%2Ftfmars/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":280901490,"owners_count":26410586,"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","status":"online","status_checked_at":"2025-10-25T02:00:06.499Z","response_time":81,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"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":["convolutional-neural-networks","human-activity-recognition","tensorflow"],"created_at":"2025-10-25T03:44:37.119Z","updated_at":"2025-10-25T03:44:42.174Z","avatar_url":"https://github.com/Shakshi3104.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# tfmars\n\n\u003cp align=\"center\"\u003e\n    \u003cimg src=\"tfmars-logo.PNG\" width=128\u003e\n\u003c/p\u003e\n\n**tfmars** is the TensorFlow's implementation of Mobile-aware Convolutional Neural Network for Sensor-based Human Activity Recognition, a sibling of [tfgarden](https://github.com/Shakshi3104/tfgarden).\nIn this repository, some CNN models implemented in tfgarden have been implemented as Attention insertable models. \nAlso, **MarNASNets** has been implemented.\n\nMARS means **M**obile-aware **A**ctivity **R**ecognition model**S**.\n\n## Models\n\n- Simple CNN: used on [the paper by Li et al](https://www.mdpi.com/1424-8220/18/2/679).\n- VGG16\n- Inception v3\n- ResNet 18\n- PyramidNet 18\n- Xception\n- DenseNet 121\n- MobileNet\n- MobileNetV2\n- MobileNetV3 Small\n- NASNet Mobile\n- MnasNet\n- EfficientNet B0\n- EfficientNet lite0\n\n### MarNASNets\n\n**MarNASNets** are the CNN architectures designed by using Bayesian-optimization Neural Architecture Search via Keras Tuner.\nMarNASNets are **mobile-aware** models that achieves higher accuracy with fewer parameters than existing models.\nThere are variations with different search spaces (A - E).\n\n## Install\n\n```bash\npip install git+https://github.com/Shakshi3104/tfmars.git\n```\n\n## Dependency\n\n- `tensorflow \u003e= 2.4.1`\n\n\n## Performance\n\n| Model | Accuracy [%] [^1] | Size [MB] [^2] | MFLOPs | Latency [ms] [^3] | CPU load [^3] |\n| :------ | :---------: | :-------: | :-----: | :----------: | :-------: |\n| Simple CNN         | 87.71    | 5.31   | 9.22    | 4.37  | 1.59      |\n| VGG16              | 89.54    | 154.00 | 357.13  | 5.64  | 1.83      |\n| Inception-v3       | 91.85    | 57.22  | 287.16  | 3.69  | 1.51      |\n| ResNet 18          | 90.53    | 15.41  | 173.72  | 2.67  | 1.50      |\n| PyramidNet 18      | 91.48    | 1.63   | 19.49   | **2.12**  | 1.65      |\n| Xception           | 92.31    | 82.69  | 613.98  | 4.09  | 2.10      |\n| DenseNet 121       | 92.55    | 22.31  | 192.97  | 2.84  | 2.11      |\n| MobileNet          | 91.22    | 23.96  | 155.47  | 2.83  | 1.88      |\n| MobileNetV2        | 90.62    | 26.91  | 147.96  | 2.96  | 1.61      |\n| MobileNetV3 Small  | 91.45    | 11.60  | 35.19   | 2.48  | **1.42**      |\n| NASNet Mobile      | 86.49    | 16.55  | 147.23  | 3.23  | 2.65      |\n| MnasNet            | 89.75    | 37.44  | 179.77  | 3.12  | 1.66      |\n| EfficientNet B0    | 92.50    | 45.70  | 221.68  | 3.32  | 1.59      |\n| EfficientNet lite0 | 91.52    | 43.11  | 220.17  | 3.21  | 1.89      |\n| MarNASNet-A        | 91.68    | 1.31   | 43.29   | 2.30  | 1.68      |\n| MarNASNet-B        | 91.79    | **0.42**   | **4.79**    | 2.21  | 1.47      |\n| MarNASNet-C        | **92.60**    | 3.08   | 46.20   | 2.22  | 1.83      |\n| MarNASNet-D        | 91.87    | 1.25   | 19.83   | 2.25  | 1.86      |\n| MarNASNet-E        | 91.70    | 8.16   | 166.26  | 2.86  | 1.46      |\n\n\n\n[^1]: Verifying accuracy with [HASC-PAC2016](http://hub.hasc.jp) (HASC).\n[^2]: Size of MLModel file.\n[^3]: Testing conducted using iPhone 12 mini with iOS 15.2. [Activitybench](https://github.com/Shakshi3104/Activitybench) 3.9.4 tested with MLComputeUnits=all. Performance tests are conducted using specific computer systems and reflect the approximate performance of iPhone 12 mini.\n\n\n## Citation \n\nUnder construction...\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fshakshi3104%2Ftfmars","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fshakshi3104%2Ftfmars","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fshakshi3104%2Ftfmars/lists"}