6-16 模型融合& WBF

模型融合方法学习总结:

https://www.cnblogs.com/libin47/p/11169994.html

 

 

Weighted Boxes Fusion: ensembling boxes for object detection models:

https://arxiv.org/abs/1910.13302

https://mp.weixin.qq.com/s?__biz=MzUxNjcxMjQxNg==&mid=2247498164&idx=3&sn=918d151c748c123e27812ef8ece23bb6&chksm=f9a18b3bced6022d9a7a0c1b9164acfebea33c437656dee662bc6fbb447cd7f902e5976e5c72&mpshare=1&scene=1&srcid=&sharer_sharetime=1587119860977&sharer_shareid=42a896371dfe6ebe8cc4cd474d9b747c&exportkey=AfuYjLAdEIfOR%2FIuYRD8dpM%3D&pass_ticket=781k4y6q5ReAogsulOWaRaDoMxACZW%2F%2FDMiqguAJb7T0aup7%2BNgkdVxCQ9fTKGHF#rd

 

WBF实例(必读):

https://www.kaggle.com/shonenkov/wbf-approach-for-ensemble(基本WBF知识)

https://www.kaggle.com/shonenkov/bayesian-optimization-wbf-efficientdet/comments(训练数据交叉分为5份分别训练得到5个模型后进行WBF)

https://www.kaggle.com/shonenkov/wbf-over-tta-single-model-efficientdet(使用TTA得到不同的预测结果后再使用TTA进行WBF)

 

https://github.com/Wakinguup/Underwater_detection(参考该代码中Weighted-Boxes-Fusion部分中的example.py可以加快理解)

××单个模型(使用不同的预测框(未进行NMS之前有多个预测框))进行WBF;多个模型的预测结果(每个模型都有一个预测框)进行WBF;

你可能感兴趣的:(6-16 模型融合& WBF)