[question] Trying to calibrate a camera and use the intrinsic parameters for reconstruction

[yan892]

Describe the problem
I have tried to calibrate a camera with checkerboard images and then change the intrinsic parameters in the CamerInit node. When I then test my dataset, the reconstructed camera positions are a bit strange. Is the calibration not good enough?

The camera is a ZED 2i stereo camera from stereolabs. (Only the images from one camera are used). It is the model with 2.12mm focal length and without polarizer.
https://cdn2.stereolabs.com/assets/datasheets/ZED%202i%20Datasheet%20Jan2023.pdf

Here is the calibration of the camera with the CameraCalibration node

Here are the intrinsic parameters of the calibration

I calculated the focal length as follows:

focal length in pixels: 1049.7
Sensor: Array Size 2688 x 1520 pixels
Pixel Size 2μm x 2μm
Image Size 2208x1242

focal length in mm = 1049.7 * 2688 * 0.002 / 2208 = 2.5558

Here I enter the intrinsic parameters in the CameraInit node:

This is what the reconstruction looks like when I specify the intrinsic parameters.
It is the default photogrammetry pipeline with imagematching method set to squential and featurematching minimal2dmotion set to 4

And this is what the reconstruction looks like if I do not specify the intrinsic parameters.

Dataset
Dataset
Chessboard images

Desktop:

• OS: win 10
• Meshroom version:
  • Binary version 2023.2.0