Opencv for Unity - Camera position inaccurate by factor of ~2

Question

I am having an issue getting accurate aruco pose estimation. The transform of the AR object relative to camera is a factor ~2 smaller than it should be. In the screenshot below, the camera is z=.40m away from the marker in real-space, but the AR object is only ~.20m away in virtual space.

I thought it might be something with the length parameters I measured off the chessboard or aruco markers?

I used these functions:

Aruco.estimatePoseSingleMarkers (corners, markerLength, camMatrix, distCoeffs, rvecs, tvecs);

where markerLength = 0.122m and is the distance across the 4x4 aruco (not including the black boarder)

    private void EstimatePoseCanonicalMarker (Mat rgbMat)
    {
        Aruco.estimatePoseSingleMarkers (corners, markerLength, camMatrix, distCoeffs, rvecs, tvecs);

        for (int i = 0; i < ids.total (); i++) {
            using (Mat rvec = new Mat (rvecs, new OpenCVForUnity.CoreModule.Rect (0, i, 1, 1)))
            using (Mat tvec = new Mat (tvecs, new OpenCVForUnity.CoreModule.Rect (0, i, 1, 1))) {
                // In this example we are processing with RGB color image, so Axis-color correspondences are X: blue, Y: green, Z: red. (Usually X: red, Y: green, Z: blue)
                Calib3d.drawFrameAxes (rgbMat, camMatrix, distCoeffs, rvec, tvec, markerLength * 0.5f);

                // This example can display the ARObject on only first detected marker.
                if (i == 0) {
                    UpdateARObjectTransform (rvec, tvec);
                }
            }
        }
    }

and:

CalcChessboardCorners (patternSize, squareSize, objectPoint, markerType);

where squareSize = .0245m and is the distance(m) between 1 square on the chessboard

    private void CalcChessboardCorners (Size patternSize, float squareSize, MatOfPoint3f corners, MarkerType markerType)
    {
        if ((int)(patternSize.width * patternSize.height) != corners.rows ()) {
            Debug.Log ("Invalid corners size.");
            corners.create ((int)(patternSize.width * patternSize.height), 1, CvType.CV_32FC3);
        }

        const int cn = 3;
        float[] cornersArr = new float[corners.rows () * cn];
        int width = (int)patternSize.width;
        int height = (int)patternSize.height;

        switch (markerType) {
        default:
        case MarkerType.ChessBoard:
        case MarkerType.CirclesGlid:
            for (int i = 0; i < height; ++i) {
                for (int j = 0; j < width; ++j) {
                    cornersArr [(i * width * cn) + (j * cn)] = j * squareSize;
                    cornersArr [(i * width * cn) + (j * cn) + 1] = i * squareSize;
                    cornersArr [(i * width * cn) + (j * cn) + 2] = 0;
                }
            }
            corners.put (0, 0, cornersArr);

            break;
        case MarkerType.AsymmetricCirclesGlid:
            for (int i = 0; i < height; ++i) {
                for (int j = 0; j < width; ++j) {
                    cornersArr [(i * width * cn) + (j * cn)] = (2 * j + i % 2) * squareSize;
                    cornersArr [(i * width * cn) + (j * cn) + 1] = i * squareSize;
                    cornersArr [(i * width * cn) + (j * cn) + 2] = 0;
                }
            }
            corners.put (0, 0, cornersArr);

            break;
        }
    }

I've tried doubling some of these values and recalibrating to see if it makes the z value close to .40m but it doesn't seem to change. Which makes me feel kind of clueless. I take about 50 calibration photos and they're pretty diverse. I try to capture a range of depth and angle on the camera.

Any ideas?

Thanks and let me know if you need more info.