Question about the calib in bev_transformer.py

[zjr-bit]

Hi, thanks for your great work!
When I read your code, I am confused about the use of calib in class sample_polar2cartin in bev_transformer.py.

translating-images-into-maps/src/model/bev_transform.py

Line 419 in bff9351

x_zmax = self.z_max / f * cu

Accoding to the meaning of intrinsics of camera, I think this should be
x_zmax = self.z_max / f * (u_max - cu)
So that we can get the maximum of X when Z is z_max.
Am I wrong? Could you please check it or let me know where my thoughts are incorrect?
Thanks in advance

[sngImagry]

I think it is fine, it is the basic formula of perspective projection: tan(fov/2)=cu/f; where f is the focal length and cu is the range between the origin_x coordinate and the center_x of the image. In the attached line we try to find a similar triangle in which the tan(fov/2)= X/self.z_max.
Therefore: X = self.z_max*cu/f

[Thang1703hrsh]

I have a question, as far as I know calib is a 3x3 matrix with 2 dimensions, but in this line of code calib is a 3-dimensional matrix f, cu = calib[:, 0, 0], calib[ :, 0, 2]

[avishkarsaha]

The first dimension accounts for the batch, so a calibration matrix per image in the batch.

…

________________________________ From: Thang1703hrsh ***@***.***> Sent: 15 June 2023 08:49 To: avishkarsaha/translating-images-into-maps ***@***.***> Cc: Subscribed ***@***.***> Subject: Re: [avishkarsaha/translating-images-into-maps] Question about the calib in bev_transformer.py (Issue #31) I have a question, as far as I know calib is a 3x3 matrix with 2 dimensions, but in this line of code calib is a 3-dimensional matrix f, cu = calib[:, 0, 0], calib[ :, 0, 2] — Reply to this email directly, view it on GitHub<#31 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AOHMZSIG6DAD4BTUVWTNLLDXLK5BTANCNFSM6AAAAAAW4XVAZ4>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[Thang1703hrsh]

So what is the calib size here, I think it's still 3x3, so when I run to that line I get the error too many indices for tensor of dimension 2