I recorded videos using a ZED stereo camera. First, I performed feature extraction using the accurate camera parameters provided by the ZED SDK. Then, I configured the rig according to the baseline length of the ZED. After that, I conducted feature matching and reconstruction. In the reconstruction results, the rotations of the two cameras in the same frame strictly follow the rig configuration, but the translation (about -0.162) is scaled up compared to the real baseline length (-0.12). This issue occurs consistently across multiple scenario tests. Why is this happening? (I have already disabled the refinement of camera parameters and rig parameters in the reconstruction settings.)

[
  {
    "cameras": [
      {
        "image_prefix": "rig1/left/",
        "ref_sensor": true,
        "camera_model_name": "PINHOLE",
        "camera_params": [524.7269897460938, 524.7269897460938, 634.0955810546875, 357.984375]
      },
      {
        "image_prefix": "rig1/right/",
        "cam_from_rig_rotation": [1.0, 0.0, 0.0, 0.0],
        "cam_from_rig_translation": [-0.12015447998046875, 0, 0],
        "camera_model_name": "PINHOLE",
        "camera_params": [524.7269897460938, 524.7269897460938, 634.0955810546875, 357.984375]
      }
    ]
  }
]