Skip to content

MetDetPy v2.1.0

Choose a tag to compare

View all tags
@Designerspr Designerspr released this 26 Aug 19:40
· 175 commits to main since this release
v2.1.0
f9b0ce0

Release of MetDetPy v2.1.0

✅ Bug Fixed

  1. 为Stacker增加出错兜底逻辑，修复Stacker在特定场景下报错会导致序列后续无输出的问题。| Added error handling logic to Stacker.
  2. 修复了2.0.1版本因为动态模板机制出错产生的效果差异。| Fixed the effect diff caused by the dynamic template mechanism error in version 2.0.1.
  3. 通过放宽重校验机制临时修复了对可能火流星的误报。会增加重校验负载。后续会对该类问题给出更稳健的解决方案。| Temporarily fixed the false positives for potential fire meteors by relaxing the revalidation mechanism.

⚠️ Revert

  1. 回滚了2.0.2版本中有关动态间隔机制的改动，维持效果一致性。该机制的下线还需要进一步测试。| Reverted the changes related to the dynamic gap mechanism in version 2.0.2 to maintain consistency. Further testing is required before deprecating this mechanism.

✅ New Feature(s)

  1. 为发行版引入 onnxruntime_directml，修复 onnxruntime 在 windows 平台不支持使用GPU的问题，缓解CPU负载。| Introduced onnxruntime_directml to the distribution, alleviating CPU load.
  2. 新增 --live-mode 选项，开启该选项可使检测速度与视频时长能够基本持平，均衡CPU负载，适用于直播场景。| Added the --live-mode option. Enabling this option allows the detection speed to match the video duration, balancing CPU load.
  3. 新增 --save 选项，允许保存运行结果到json文件。| Added the --save option, allowing the saving of results to a JSON file.

⚠️ 提供了 --provider cpu 以指定CPU运行模型，如果在未安装 DirectX 12的电脑上使用 onnxruntime_directml 造成报错，可以作为降级方案。| If using onnxruntime_directml on a computer without DirectX 12 installed causes errors, use --provider cpu option to specify CPU as a fallback solution.

✅ Improvement(s)

  1. 更新了评估工具evaluate.py，更改了评估标注格式，支持保存结果，与其他版本比较性能和效果等。该结果可作为回归测试与性能测试的基准，以确保迭代间的性能和效果稳定性。| Updated the evaluation tool evaluate.py, changing the evaluation annotation format to support saving results and comparing performance. These results can serve as a benchmark for regression testing and performance testing.

Full Changelog: v2.0.2...v2.1.0

Assets 4
Loading