Transformation of relative pose to absolute pose

[lucleon]

In line 94 in test.py, the Rotation Matrix is computed using only one Euler angle:

ang = eulerAnglesToRotationMatrix([0, answer[-1][0], 0])

I'm wondering, why are theta_x and theta_z set to zero?

[fangxu622]

@ChiWeiHsiao @lucleon Hi , I have same questions.

[yangxiaoliyang]

I have the same problem, and when I evaluate the error of relative pose, the error is very large, but the trajectory fits well. Have you solved it? @fangxu622 @lucleon @ChiWeiHsiao

[mrdavoodi]

KITTI is recorded from the trajectory of a car. This movement does not have pitch and roll. The only angle change is yaw.

[ztysdu]

I have the same problem, and when I evaluate the error of relative pose, the error is very large, but the trajectory fits well. Have you solved it? @fangxu622 @lucleon @ChiWeiHsiao

In my opinion,This is due to in test.py,they measure the loss in absolute pose,which is different from training

[yangxiaoliyang]

It's worked out. This is the absolute pose. It can be solved by processing the pose data in the data set as relative pose.Hope to help you!

…

------------------ 原始邮件 ------------------ 发件人: "ChiWeiHsiao/DeepVO-pytorch" ***@***.***>; 发送时间: 2022年11月22日(星期二) 上午9:40 ***@***.***>; ***@***.******@***.***>; 主题: Re: [ChiWeiHsiao/DeepVO-pytorch] Transformation of relative pose to absolute pose (#30) I have the same problem, and when I evaluate the error of relative pose, the error is very large, but the trajectory fits well. Have you solved it? @fangxu622 @lucleon @ChiWeiHsiao In my opinion,This is due to in test.py,they measure the loss in absolute pose,which is different from training — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>

[yangxiaoliyang]

Thank you for your reply！

…

------------------ 原始邮件 ------------------ 发件人: "ChiWeiHsiao/DeepVO-pytorch" ***@***.***>; 发送时间: 2022年11月22日(星期二) 上午9:40 ***@***.***>; ***@***.******@***.***>; 主题: Re: [ChiWeiHsiao/DeepVO-pytorch] Transformation of relative pose to absolute pose (#30) I have the same problem, and when I evaluate the error of relative pose, the error is very large, but the trajectory fits well. Have you solved it? @fangxu622 @lucleon @ChiWeiHsiao In my opinion,This is due to in test.py,they measure the loss in absolute pose,which is different from training — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>