Skip to content

Releases

Wulabel v0.6.0

Choose a tag to compare

@github-actions github-actions released this 21 Nov 11:00

更新内容

  • 实时验证和推理服务包支持更多模型:分割、关键点、分类模型都能实时验证、生成推理服务包;视频流支持 RTSP、HTTP-FLV、HLS(m3u8)和 MP4。
  • 训练集划分看得见、能调整:训练页「查看和调整」里能看到每张图分在训练集、验证集还是测试集,能手动移动,也能按比例重新划分。
  • 一个人标注不用审核:标了目标就算已标注,确实没有目标的图点「完成」作为负样本;团队协作时才有审核流程。
  • 训练不再白跑:训练前提示每轮和总共的步数,数据太少会先提醒;训练中指标一直是 0、没下载到预训练权重时醒目提示;常见报错直接给出解决办法;训练完成的通知带上结果。
  • 评估给结论:评估页给出「差 / 一般 / 可用」的结论和最差的类别,推荐的报警阈值自动带到部署页。
  • 标注更顺手:标完一批有提示,「完成」直接跳到下一张未标注;新增 X 跳过、Shift+D 下一张未标注;底部缩略图显示状态,可以输入序号跳转;批量预标注可以中途停止;删除类别时可以合并到另一个类别。
  • 数据页批量操作:批量设为跳过、恢复未标注、送去标注、移到训练/验证/测试集;视频抽帧和导入图片可以中途停止;把视频拖进窗口直接抽帧。
  • 体检更安全:去重时优先保留有标注的图,也可以自己选留哪张;忽略的问题可以恢复,疑似漏标可以一次全部加上。
  • 其他:新建项目时可以直接填类别;预训练权重下载不完整时自动删掉重下。

What's new (English)

  • Live verification and the inference service package support segmentation, keypoint and classification models; video streams support RTSP, HTTP-FLV, HLS (m3u8) and MP4.
  • The Train page shows which images are in the training, validation and test sets; move images between sets or re-split by ratio.
  • No review needed when labeling on your own; the review workflow is only used in team collaboration.
  • Training shows steps per epoch and in total and warns about too little data; clear warnings when the metric stays at 0 or pretrained weights couldn't be downloaded; common errors come with a fix.
  • The Evaluate page gives a verdict and the worst class; the recommended alarm threshold carries over to the Deploy page.
  • Annotate: end-of-batch prompt, "Done" jumps to the next unlabeled image, new shortcuts X (skip) and Shift+D (next unlabeled), status dots on thumbnails, jump by number, stop batch pre-labeling, merge a class into another when deleting it.
  • Data: bulk skip / restore / send to annotation / move to a split; frame extraction and image import can be stopped; drag a video in to extract frames.
  • Health check keeps labeled images when removing duplicates; ignored issues can be restored; add all suspected missing labels at once.

下载 / Download

系统 / System 文件 / File
Windows 10 / 11(64 位 / 64-bit) Wulabel_0.6.0_x64-setup.exe
macOS 11+(Apple 芯片和 Intel 通用 / Apple silicon & Intel) Wulabel_0.6.0_universal.dmg
Linux x86_64 Wulabel_0.6.0_amd64.deb / Wulabel_0.6.0_amd64.AppImage
Linux ARM64 Wulabel_0.6.0_arm64.deb / Wulabel_0.6.0_aarch64.AppImage

国内下载更快 / Website: https://wulabel.xiaowuiot.com

带 .sig、.tar.gz 的文件是软件自动更新用的,手动安装不用下载。Files ending in .sig or .tar.gz are used by the in-app updater.

有问题可以在软件右上角「反馈」,或发邮件到 177183719@qq.com。Questions: Feedback in the app, 177183719@qq.com, or GitHub Issues.

Wulabel v0.5.2

Choose a tag to compare

@github-actions github-actions released this 20 Nov 07:47

更新内容

  • 国际版改进:「看使用教程」和更新提示里的官网链接,按界面语言打开对应语言的页面;反馈、检查更新时服务器返回的提示也跟随界面语言。
  • 巡瞳 AI 开关:巡瞳 AI 是国内平台,非中文界面默认隐藏相关功能,需要的话在「设置 → 巡瞳 AI」里打开。简体中文界面不受影响。
  • 译文修正:越南语、俄语、印尼语、西班牙语、德语里几处不通顺或前后不一致的译文。
  • Linux 自动下载 ffmpeg:视频抽帧时自动下载,不用再手动安装(x86_64 和 ARM64 都支持)。

What's new (English)

  • The "View tutorial" button and the website links in update notices open the page in your interface language; server messages for feedback and update checks follow the interface language too.
  • Xuntong AI (a platform in China) is hidden by default in non-Chinese interfaces; turn it on in Settings → Xuntong AI if you need it.
  • Fixed awkward or inconsistent translations in Vietnamese, Russian, Indonesian, Spanish and German.
  • On Linux, ffmpeg is now downloaded automatically when extracting frames from videos.

下载 / Download

系统 / System 文件 / File
Windows 10 / 11(64 位 / 64-bit) Wulabel_0.5.2_x64-setup.exe
macOS 11+(Apple 芯片和 Intel 通用 / Apple silicon & Intel) Wulabel_0.5.2_universal.dmg
Linux x86_64 Wulabel_0.5.2_amd64.deb / Wulabel_0.5.2_amd64.AppImage
Linux ARM64 Wulabel_0.5.2_arm64.deb / Wulabel_0.5.2_aarch64.AppImage

国内下载更快 / Website: https://wulabel.xiaowuiot.com

带 .sig、.tar.gz 的文件是软件自动更新用的,手动安装不用下载。Files ending in .sig or .tar.gz are used by the in-app updater.

有问题可以在软件右上角「反馈」,或发邮件到 177183719@qq.com。Questions: Feedback in the app, 177183719@qq.com, or GitHub Issues.

Wulabel v0.5.1

Choose a tag to compare

@github-actions github-actions released this 20 Nov 06:48

更新内容

  • 浅色主题:默认跟随系统的浅色或深色模式,也可以在「设置 → 外观」里手动切换。标注画布、图片预览等看图的区域仍保持深色,看图更准。
  • 窄窗口和外文界面适配:窗口较窄或使用英文等较长的语言时,首页新手引导自动排成多行,左侧导航的长词也能完整显示。

v0.4.1 起点「立即更新」就是全自动的;更早的版本请到官网下载,直接覆盖安装。

What's new (English)

  • Light theme: follows your system's light or dark mode by default; you can also switch it in Settings → Appearance. Image areas such as the annotation canvas stay dark so images are easier to judge.
  • Better layout in narrow windows and other languages: the home page guide wraps onto several rows, and long sidebar labels fit.

下载 / Download

系统 / System 文件 / File
Windows 10 / 11(64 位 / 64-bit) Wulabel_0.5.1_x64-setup.exe
macOS 11+(Apple 芯片和 Intel 通用 / Apple silicon & Intel) Wulabel_0.5.1_universal.dmg
Linux x86_64 Wulabel_0.5.1_amd64.deb / Wulabel_0.5.1_amd64.AppImage
Linux ARM64 Wulabel_0.5.1_arm64.deb / Wulabel_0.5.1_aarch64.AppImage

国内下载更快 / Website: https://wulabel.xiaowuiot.com

带 .sig、.tar.gz 的文件是软件自动更新用的,手动安装不用下载。Files ending in .sig or .tar.gz are used by the in-app updater.

有问题可以在软件右上角「反馈」,或发邮件到 177183719@qq.com。Questions: Feedback in the app, 177183719@qq.com, or GitHub Issues.

Wulabel v0.5.0

Choose a tag to compare

@github-actions github-actions released this 20 Nov 05:56

更新内容

国际版:支持 12 种语言

  • 界面支持简体中文、繁體中文、English、日本語、한국어、Español、Português、Русский、Deutsch、Français、Tiếng Việt、Bahasa Indonesia。
  • 默认跟随系统语言(中文系统显示中文,没有对应语言的显示英文),也可以在「设置 → 语言」里手动切换。
  • 出错提示、训练日志、交付报告、推理服务包的说明和启动提示,也按界面语言显示。
  • 官网和使用教程也有了各语言版本:https://wulabel.xiaowuiot.com
  • 译文由 AI 翻译,发现不准确的地方欢迎点右上角「反馈」告诉我们。

v0.4.1 起点「立即更新」就是全自动的;v0.4.0 升级会下载并打开安装程序;v0.3.0 及更早的版本请到官网下载,直接覆盖安装。

What's new (English)

  • International edition: the interface is now available in 12 languages and follows your system language by default (English for languages that aren't included). You can change it in Settings → Language.
  • Error messages, training logs, delivery reports and the inference service package's instructions also use the interface language.
  • The website and user guide are available in these languages too: https://wulabel.xiaowuiot.com/en/
  • Translations are done by AI. If you spot something wrong, please tell us via Feedback in the top-right corner or open an issue.

下载 / Download

系统 / System 文件 / File
Windows 10 / 11(64 位 / 64-bit) Wulabel_0.5.0_x64-setup.exe
macOS 11+(Apple 芯片和 Intel 通用 / Apple silicon & Intel) Wulabel_0.5.0_universal.dmg
Linux x86_64 Wulabel_0.5.0_amd64.deb(sudo apt install ./Wulabel_0.5.0_amd64.deb)/ Wulabel_0.5.0_amd64.AppImage
Linux ARM64 Wulabel_0.5.0_arm64.deb / Wulabel_0.5.0_aarch64.AppImage

带 .sig、.tar.gz 的文件是软件自动更新用的,手动安装不用下载。Files ending in .sig or .tar.gz are used by the in-app updater.

第一次安装 / First install

  • Windows 弹出「Windows 已保护你的电脑」:点「更多信息」→「仍要运行」。If you see "Windows protected your PC", click "More info" → "Run anyway".
  • Mac 提示「无法验证开发者」:在「应用程序」里右键 Wulabel →「打开」;提示「已损坏」的话,在终端执行 xattr -cr /Applications/Wulabel.app。On macOS, right-click Wulabel in Applications → "Open"; if it says the app is damaged, run xattr -cr /Applications/Wulabel.app.

有问题可以在软件右上角「反馈」,或发邮件到 177183719@qq.com。Questions: Feedback in the app, 177183719@qq.com, or GitHub Issues.

Wulabel v0.4.2

Choose a tag to compare

@github-actions github-actions released this 20 Nov 01:27

更新内容

新增 ARM64 Linux 版

  • 飞腾、鲲鹏等 ARM 电脑上的银河麒麟、统信 UOS 等系统可以用了,有 .deb 和 .AppImage 两种。
  • 训练环境会自动装 ARM 版 PyTorch。

训练显存不够自动重试

  • 训练时显存不够不再直接失败:软件自动把批大小减半、从头重新训练,界面上会提示,训练记录里写明实际用的批大小。

更新后显示这一版改了什么

  • 自动更新装好、重新打开软件时,会弹出这一版的更新内容,看过一次就不再弹。

反馈可以附带运行日志

  • 右上角「反馈」默认附带训练环境、最近两次训练的日志和出错信息,方便排查问题;可以先点「看看内容」确认,不想带就取消勾选。不含图片和标注。

其他

  • 用 .deb 安装的 Linux 版也能软件内全自动更新了。
  • 修复 RTSP 视频流抽帧:超时设置对 RTSP 不生效、连不上时会一直卡住;TCP 连不上会自动改用 UDP 再试;账号密码不对、地址路径不对会直接说明原因。

v0.4.1 起点「立即更新」就是全自动的;v0.4.0 升级会下载并打开安装程序;v0.3.0 及更早的版本请到官网下载,直接覆盖安装。

下载

系统 文件
Windows 10 / 11(64 位) Wulabel_0.4.2_x64-setup.exe
macOS 11 及以上(Apple 芯片和 Intel 通用) Wulabel_0.4.2_universal.dmg
Linux x86_64 Wulabel_0.4.2_amd64.deb(sudo apt install ./Wulabel_0.4.2_amd64.deb)或 Wulabel_0.4.2_amd64.AppImage(chmod +x 后直接运行)
Linux ARM64(飞腾、鲲鹏) Wulabel_0.4.2_arm64.deb 或 Wulabel_0.4.2_aarch64.AppImage

国内下载更快:https://wulabel.xiaowuiot.com

带 .sig、.tar.gz 的文件是软件自动更新用的,手动安装不用下载。

第一次安装

  • Windows 弹出「Windows 已保护你的电脑」:点「更多信息」→「仍要运行」。
  • Mac 提示「无法验证开发者」:在「应用程序」里右键 Wulabel →「打开」;提示「已损坏」的话,在终端执行 xattr -cr /Applications/Wulabel.app。

有问题可以在软件右上角「反馈」,或发邮件到 177183719@qq.com。

Wulabel v0.4.1

Choose a tag to compare

@github-actions github-actions released this 19 Nov 11:16

更新内容

全自动更新

  • 有新版本时点「立即更新」,软件在后台下载、校验签名、安装,装好后自动重新打开,不用再手动运行安装程序。
  • Windows 安装时只显示一个进度条,不用点「下一步」;Mac 直接替换后重启。项目和数据都会保留。
  • 更新包带数字签名,被替换或没下载完整的包不会被安装。

全自动更新从 v0.4.1 开始才有:v0.4.0 升级到这个版本时,还是会下载并打开安装程序,按提示装好即可;以后的版本就是全自动了。v0.3.0 及更早的版本请到官网下载,直接覆盖安装。

下载

系统 文件
Windows 10 / 11(64 位) Wulabel_0.4.1_x64-setup.exe
macOS 11 及以上(Apple 芯片和 Intel 通用) Wulabel_0.4.1_universal.dmg
Linux x86_64 Wulabel_0.4.1_amd64.deb(sudo apt install ./Wulabel_0.4.1_amd64.deb)或 Wulabel_0.4.1_amd64.AppImage(chmod +x 后直接运行)

国内下载更快:https://wulabel.xiaowuiot.com

带 .sig、.tar.gz 的文件是软件自动更新用的,手动安装不用下载。

第一次安装

  • Windows 弹出「Windows 已保护你的电脑」:点「更多信息」→「仍要运行」。
  • Mac 提示「无法验证开发者」:在「应用程序」里右键 Wulabel →「打开」;提示「已损坏」的话,在终端执行 xattr -cr /Applications/Wulabel.app。

Wulabel v0.4.0

Choose a tag to compare

@github-actions github-actions released this 19 Nov 10:00

更新内容

新增 Linux 版

  • 支持 Linux x86_64:Ubuntu 22.04 及以上、Debian 12、统信 UOS、x86 电脑上的银河麒麟 / 中标麒麟等,提供 .deb 和免安装的 .AppImage。

更好上手

  • 首页新手引导:装训练环境 → 新建项目 → 导入图片 → 标一小批 → 训练 → 预标注和部署,做完的步骤自动打勾。
  • 项目卡片显示封面、图片数、标注进度和训练次数。
  • 软件内一键升级:有新版本时点「立即更新」,自动下载、校验并打开安装程序。

训练

  • 检测项目默认用 YOLOX-S,按这台电脑的显卡推荐规格(没有独立显卡 Tiny、4GB 以上 S、8GB 以上 M、Mac S)。
  • 训练结束发系统通知,可以直接「去评估」「去预标注」。
  • 训练页常用设置和高级设置分开,高级设置默认收起。

标注

  • 文本提示:中文类别名自动对应英文(安全帽 → helmet,内置 160 多个常见词),不用再手动填英文提示词。
  • 标注页按 ? 打开快捷键面板。

部署

  • 分割、关键点模型(Mask R-CNN、Keypoint R-CNN)也能导出 ONNX、OpenVINO、TensorRT / RKNN 转换包。

反馈

  • 反馈时可以留邮箱,问题处理好了作者会发邮件告诉你。

下载

系统 文件
Windows 10 / 11(64 位) Wulabel_0.4.0_x64-setup.exe
macOS 11 及以上(Apple 芯片和 Intel 通用) Wulabel_0.4.0_universal.dmg
Linux x86_64 Wulabel_0.4.0_amd64.deb(sudo apt install ./Wulabel_0.4.0_amd64.deb)或 Wulabel_0.4.0_amd64.AppImage(chmod +x 后直接运行)

国内下载更快:https://wulabel.xiaowuiot.com

暂不支持:ARM 芯片的 Linux(飞腾、鲲鹏,计划中)、龙芯、鸿蒙。

第一次安装

  • Windows 弹出「Windows 已保护你的电脑」:点「更多信息」→「仍要运行」。
  • Mac 提示「无法验证开发者」:在「应用程序」里右键 Wulabel →「打开」;提示「已损坏」的话,在终端执行 xattr -cr /Applications/Wulabel.app。

Wulabel v0.3.0

Choose a tag to compare

@github-actions github-actions released this 17 Nov 07:20

更新内容

新增 YOLO 系列训练(YOLOX)

  • 训练页新增 YOLOX-Nano / Tiny / S / M / L 五种规格:Nano、Tiny 适合 CPU 和 NPU 边缘盒子,S 速度和精度最均衡,M、L 适合有显卡、要精度的场景。
  • 自带官方 COCO 预训练权重(第一次训练时自动下载,国内从官网镜像下载)。
  • 按 YOLOX 官方方法训练:Mosaic、MixUp、随机缩放旋转、颜色抖动、多尺度训练、EMA,最后几轮自动关掉强增强收尾。建议训 50 轮以上。
  • 训好的 YOLOX 模型和其他模型一样用:评估(mAP、混淆矩阵、阈值推荐)、模型预标注、误报测试、实时验证、接着训练。
  • 部署:ONNX / OpenVINO 和推理服务包直接可用,也能上传巡瞳 AI 平台;固定尺寸 ONNX、TensorRT、RKNN 转换包用 YOLOX 官方格式(模型里没有 NMS,NPU 和 TensorRT 都能完整跑)。
  • YOLOX 是 Apache-2.0 协议,训出来的模型可以放心商用(YOLOv5 / v8 / v11 是 AGPL-3.0,所以没有采用)。第三方代码说明见仓库的 NOTICE。

下载

  • Windows 10 / 11(64 位):Wulabel_0.3.0_x64-setup.exe
  • macOS 11 及以上(Apple 芯片和 Intel 通用):Wulabel_0.3.0_universal.dmg

国内下载更快:https://wulabel.xiaowuiot.com

装了 v0.2.1 的用户,启动软件时会自动提示更新。

第一次安装

  • Windows 弹出「Windows 已保护你的电脑」:点「更多信息」→「仍要运行」。
  • Mac 提示「无法验证开发者」:在「应用程序」里右键 Wulabel →「打开」;提示「已损坏」的话,在终端执行 xattr -cr /Applications/Wulabel.app。

Wulabel v0.2.1

Choose a tag to compare

@github-actions github-actions released this 12 Nov 06:30

更新内容

  • 新增检查更新:启动时检查有没有新版本,有的话右下角提醒;设置页「关于与更新」里可以手动检查,也可以关掉自动检查。检查时只发送版本号、系统类型和一个随机的匿名编号,不含图片和个人信息。
  • 缩略图条可以调大小:标注页底部的缩略图条,拖它的上边缘调整大小,双击恢复默认。
  • 安装界面改成中文:Windows 安装程序改为简体中文。

下载

  • Windows 10 / 11(64 位):Wulabel_0.2.1_x64-setup.exe
  • macOS 11 及以上(Apple 芯片和 Intel 通用):Wulabel_0.2.1_universal.dmg

国内下载更快:https://wulabel.xiaowuiot.com

第一次安装

  • Windows 弹出「Windows 已保护你的电脑」:点「更多信息」→「仍要运行」。
  • Mac 提示「无法验证开发者」:在「应用程序」里右键 Wulabel →「打开」;提示「已损坏」的话,在终端执行 xattr -cr /Applications/Wulabel.app。

Wulabel v0.2.0

Choose a tag to compare

@github-actions github-actions released this 10 Nov 16:01

离线 AI 标注训练桌面软件:导入数据 → 标注 → 训练 → 评估 → 模型预标注 → 导出部署,全流程在本机完成,数据不出本机。

下载

系统 文件
Windows 10 / 11(64 位) Wulabel_0.2.0_x64-setup.exe,双击安装
macOS 11 及以上(Apple 芯片和 Intel 通用) Wulabel_0.2.0_universal.dmg,打开后把 Wulabel 拖进「应用程序」

国内下载慢的话,到官网下载:https://wulabel.xiaowuiot.com

Mac 第一次打开:软件没有苹果开发者签名,会提示「无法验证开发者」。在「应用程序」里右键 Wulabel → 打开 → 再点「打开」;macOS 15 以后先双击一次,再到「系统设置 → 隐私与安全性」点「仍要打开」。提示「已损坏」的话,在终端执行 xattr -cr /Applications/Wulabel.app。只需要第一次这样做。

新功能

  • 四种任务:目标检测之外,新增实例分割、图像分类、关键点(人体姿态等),都能标注、训练、评估、预标注。
  • AI 辅助标注:SAM 点一下出轮廓;按类别名自动找目标;视频关键帧插值、往后自动跟踪。
  • 实时多人协作:你的电脑当主机,成员用浏览器登录就能一起标,不用装软件;局域网或在线都行。分配任务、按张加锁、审核打回、工作量统计、水印。
  • 巡瞳 AI 对接:登录巡瞳账号即可使用;按项目类别识别、和 AI 聊天定识别规则、专用模型批量预标注;模型一键上传到平台,告警图拉回来标注。
  • 数据:视频抽帧、RTSP / 直播流抽帧、图片质量筛查、主动学习挑图、数据集版本、导出时人脸 / 车牌打码。
  • 训练与评估:RetinaNet / FCOS / SSDLite、Mask / Keypoint R-CNN、分类模型;数据增强、继续训练、训练对比、导入外部 ONNX;阈值推荐、误报测试、交付报告。
  • macOS 版:Apple 芯片训练自动用 MPS 加速。

使用教程:https://wulabel.xiaowuiot.com/guide.html
Apache-2.0 开源。