Skip to content

Vision Workbench v0.3.0

Choose a tag to compare

View all tags
@ksukie ksukie released this 13 Jul 10:43
· 13 commits to main since this release
v0.3.0
7f58498

Vision Workbench v0.3.0

English

Vision Workbench v0.3.0 strengthens model and dependency safety while making the desktop workflows easier to try immediately.

Highlights

  • Safer model loading and downloads: restricted checkpoint loading, HTTPS/local URL rules, size limits, checksum validation, and archive-expansion limits.
  • Added model-manifest refresh, single-image detection API support, and best-weight registration after training.
  • Training pages now include deterministic sample datasets, runtime diagnostics, recommended batch sizes, clearer beginner guidance, and stoppable classification training with epoch metrics.
  • Added a bundled sample image with Load Sample Image controls across image-input desktop workflows for quick testing.
  • Improved keyboard navigation, focus behavior, labels, and assistive-technology metadata.
  • CI now uses hash-locked dependencies, dependency auditing, CycloneDX SBOM generation, and broader Python/platform coverage.
  • Updated Pillow to 12.3.0 to address dependency security findings.

Compatibility

The default safety policy may reject untrusted model URLs, oversized inputs, unsafe training run names, and conflicting user-site packages in isolated environments.

Install

pip install vision_workbench-0.3.0-py3-none-any.whl
vision-workbench

中文

Vision Workbench v0.3.0 重点增强了模型与依赖安全性，并让桌面端功能可以更方便地直接试用。

主要更新

  • 更安全的模型加载与下载：受限 checkpoint 加载、HTTPS/本地 URL 限制、文件大小上限、SHA-256 校验与压缩包解压限制。
  • 新增模型清单刷新、单图目标检测 API，以及训练完成后注册最佳权重。
  • 训练页面新增确定性示例数据、运行环境诊断、推荐批量大小、新手参数提示，以及可停止并显示每轮指标的分类训练。
  • 所有接收输入图像的桌面流程新增内置示例图和“加载示例图”按钮，便于快速测试。
  • 改进键盘导航、焦点状态、表单标签与辅助技术支持。
  • CI 新增带哈希依赖锁定、安全审计、CycloneDX SBOM，以及更广泛的 Python 与平台覆盖。
  • Pillow 升级至 12.3.0，修复依赖安全审计发现的问题。

兼容性说明

默认安全策略可能拒绝不受信任的模型 URL、超限输入、不安全的训练运行名，以及隔离环境中冲突的用户级包。

安装

pip install vision_workbench-0.3.0-py3-none-any.whl
vision-workbench
Assets 5
Loading