单目深度估计评估指标

KITTI Depth以及ScanNet评估指标

 

指标 名称 表达式
abs rel. absolute relative error abs\ rel. = \frac{1}{n} \sum | \frac{y_{pred} - y_{gt}}{y_{gt}} |
mae mean absolute error MAE =\frac{1}{n} \sum | y_{pred} - y_{gt} |
log mae mean absolute logarithmic error log\ MAE=\frac{1}{n} \sum | log(y_{pred}) - log(y_{gt}) |
imae inverse mean absolute error iMAE =\frac{1}{n} \sum | \frac{1}{y_{pred}} - \frac{1}{y_{gt}} |
rmse root mean square error RMSE = \sqrt{\frac{1}{n} \sum ({y_{pred}} - {y_{gt}})^2}
log rmse root mean square logarithmic error log\ RMSE = \sqrt{\frac{1}{n} \sum (log(y_{pred}) - log(y_{gt}))^2}
irmse inverse root mean square error iRMSE = \sqrt{\frac{1}{n} \sum (\frac{1}{y_{pred}} - \frac{1}{y_{gt}})^2}

sq. rel.

square relative error sq.\ rel. = \frac{1}{n} \sum (\frac{y_{pred} - y_{gt}}{y_{gt}} ) ^2
scale invar. (SILog) scale invariant logarithmic error SILog=\frac{1}{n} \sum_{i}^{n} d_i^2 + \frac{1}{n^2} (\sum_i^n d_i)^2 \ \ \ \ \ \ \ \ d = log(y_{pred}) - log(y_{gt})

 

你可能感兴趣的:(单目深度估计)