第二种DTU的评估方式

MiZhenxing / GBi-Net

Codes for GBi-Net (CVPR2022)

https://mizhenxing.github.io/gbinet

MIT License

124 stars 15 forks source link

Closed KyrieLiu52 closed 2 years ago

KyrieLiu52 commented 2 years ago

您好您的论文中提到对DTU数据集使用了两种评价指标，第二种是直接评估深度图预测的准确性。请问这部分是您自己提出的一种评价方法，还是类似第一种基于距离的评价指标，由其他方法（如DTU）提供的评估方式。谢谢！

MiZhenxing commented 2 years ago

Hi, the second evaluation metric directly subtracts the predicted depth maps with the gt depth maps and gets the absolute error. Then we set a threshold and compute the percentage of pixels whose error is below the threshold. You could reference the code in MVSNet-pytorch. https://github.com/xy-guo/MVSNet_pytorch/blob/e0f2ae3d7cb2dd13807b775f2075682eaa7f1521/utils.py#L139-L146

KyrieLiu52 commented 2 years ago

非常感谢您的回复