Skip to content

MetDetPy v2.2.0

Latest
Latest
Compare
Choose a tag to compare
@Designerspr Designerspr released this 10 Jan 11:07
v2.2.0
05d89cf

Release Version of MetDetPy v2.2.0

（鸽了一年千呼万唤始出来的）MetDetPy v2.2.0 版本🌚🌚。主要更新有：
MetDetPy V2.2.0, a long-delayed version 🌚🌚(my fault). Main updates are:

✅ New Feature(s)

  1. 更新默认检测模型，优化检测稳定性，支持了更多检测类别。| Update the default detection model , improve the robustness and support more detection categories.
  2. 新增图像检测的工具入口 MetDetPhoto，可用于单张照片，延时序列（照片文件夹）或者延时序列视频的检测。 | Release MetDetPhoto for image meteor detection, which can be used for single photo, time-lapse image sequence and video.
  3. 检测模型支持多尺度推理，改善了流星召回率。| Model now supports multi-scale inference, improving target recall.
  4. 正式定义流星检测结果格式(MDRF)作为MetDetPy的流转文件格式，增加各脚本的通用性。有关该脚本的更多介绍见工具文档。| Introduce meteor detection result format (MDRF) as the standard file format of MetDetPy to increase the versatility of scripts. For more information about this script, see Tool Usage.

✅ Improvement(s)

  1. 调整了重校验阈值与前置预测的得分的融合计算方式，改善召回性能。| Adjust the fusion calculation method of the recheck threshold and the final predicted score, and improve the recall performance.
  2. 改进了 ClipToolkit 的易用性，提供了简化的调用接口，并允许同时保存Labelme格式标注或保存带标注框的图像等。用法见文档。| Improve the usability of ClipToolkit, providing a simplified version API, and allowing to save Labelme format annotations or images with annotation boxes. See the documents for more information.
  3. 增加了速度，距离和持续时间的归一化统计值输出，为不同设备上的统计值范围提供了统一参考。| Add the output of normalized speed, distance, and duration statistics, providing a unified reference for statistical uses.

✅ Bug Fixed

  1. 修复流星置信判定时速度统计值未归一化的问题，改善对卫星和飞机的误判。| Use normalized speed statistics to determine meteor confidence intervals, expecting to improve misjudgment of satellites and aircraft.

⚠️ Modification(s)

  1. 调整了后端模式下的输出逻辑，后端现在会输出所有检测类别（同样以 Meteor 开头）。| Adjust the output logic in the backend mode. All detection results will be output in the backend (starting with Meteor).
  2. 调整了 evaluate 的参数解析逻辑以适配MDRF格式。| Adjust the parsing logic of evaluate.py to support MDRF.
  3. 调整了保存到文件的结果，使其与命令行输出流一致。| Unify the coordinates and format saved to the test result file with the command line output.

⚠️

  1. 本次更新对默认参数范围有较大变动，使用从先前版本复制的自定义配置文件可能在新版本不适用，建议用户参考参数解释页手动更新配置文件。| This update has major changes to the default parameter ranges. If you want to use custom configuration files from previous versions, it is recommended to refer to the parameter explanation page to manually update the configuration files.
  2. 本次模型的更新对图像中目标的分类可能与先前版本存在Diff。如果希望沿用先前版本的模型，请修改模型相关路径为旧版本模型。| v2 model update may diff from the results of the previous version. If you want to use the previous version of the model, please modify the model-related path to the old version of the model.
  3. 由于稀有精灵样本较少，模型对该类别的识别准召率暂不理想，暂时不建议启用"RARE_SPRITE"作为正样本类型。| Due to the limited number of rare sprite samples, the model's precision and recall for this category are not ideal now. It is not recommended to enable "RARE_SPRITE" as positive category for now.

详细更新内容见更新日志-v2.2.0。

Full Changelog: v2.1.0...v2.2.0

Assets 4
Loading