{"id":21085138,"url":"https://github.com/wmpscc/scenerecognize","last_synced_at":"2025-07-09T14:36:10.766Z","repository":{"id":141563481,"uuid":"135886336","full_name":"wmpscc/SceneRecognize","owner":"wmpscc","description":"AI Challenger: 对80个场景进行分类","archived":false,"fork":false,"pushed_at":"2018-06-05T07:51:20.000Z","size":11,"stargazers_count":4,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"master","last_synced_at":"2025-03-14T05:13:33.054Z","etag":null,"topics":["ai","challenger","scene-classification"],"latest_commit_sha":null,"homepage":null,"language":"Python","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/wmpscc.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":"2018-06-03T08:22:39.000Z","updated_at":"2020-04-27T17:52:48.000Z","dependencies_parsed_at":null,"dependency_job_id":"2df22703-698b-4c68-9a7b-34529424e357","html_url":"https://github.com/wmpscc/SceneRecognize","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/wmpscc/SceneRecognize","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/wmpscc%2FSceneRecognize","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/wmpscc%2FSceneRecognize/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/wmpscc%2FSceneRecognize/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/wmpscc%2FSceneRecognize/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/wmpscc","download_url":"https://codeload.github.com/wmpscc/SceneRecognize/tar.gz/refs/heads/master","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/wmpscc%2FSceneRecognize/sbom","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":264476614,"owners_count":23614542,"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":["ai","challenger","scene-classification"],"created_at":"2024-11-19T20:29:57.912Z","updated_at":"2025-07-09T14:36:10.702Z","avatar_url":"https://github.com/wmpscc.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# SceneRecognize\n[AI Challenger](https://challenger.ai/competition/scene/subject): 对80个场景进行分类\n\n# 模型实现\n- 使用TensorFlow 的slim模块实现网络的基础结构。\n- 构建深度残差网络。\n# 模型特点\n模型设计主要参考了VGG16和Inception-ResNet-V2的结果，并有所改进。\u003cbr\u003e\n- 在制作TFRecord时，将输入数据尺寸统一为299x299x3\n- 将激活函数变为Leaky relu 增加模型非线性能力。\n- 加入BN层，可选择去掉dropout，增加模型稳定性，减少参数量\n- 使用Global Pooling层替换传统的全连接dense层。\n- 使用核为1x1的Conv2D替换全连接层。\n- 调整网络深度，以适应当前数据集\n\n# 设计原则\n- 对于相同的尺寸的输出特征图谱，每层必须含有相同数量的过滤器\n- 如果特征图谱的尺寸减半，则过滤器的数量必须翻倍，以保持每层的时间复杂度\n- 直接通过卷积层（stride=2）进行下采样，网络末端以全局的均值池化层结束\n\n# 关键代码\n``` Python\nnet = slim.batch_norm(net, is_training=True)\n```\n``` Python\n# Global Average Pooling\n    kernel_size = net.get_shape()[1:3]\n    if kernel_size.is_fully_defined():\n        net = slim.avg_pool2d(net, kernel_size, padding='VALID')\n    else:\n        net = tf.reduce_mean(net, [1, 2], keep_dims=True)\n\n    # dropout\n    if dropout_keep_prob != 1:\n        net = slim.dropout(net, dropout_keep_prob, is_training=True)\n\n    # Use conv2d instead of fully_connected layers.\n    net = slim.conv2d(net, 80, [1, 1])\n```\n\n# 参考文献\n- [图像识别领域的深度残差学习](https://www.jianshu.com/p/f71ba99157c7)\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fwmpscc%2Fscenerecognize","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fwmpscc%2Fscenerecognize","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fwmpscc%2Fscenerecognize/lists"}