Bug in TestRingRectification.cpp

[PaulBergmann]

In TestRingRectification.cpp (Line 87), you have the following for loop for reading the side camera images for different frames:

vector<vector<Mat>> sideCamImages(framesList.size(), vector<Mat>());
  for (int frameIdx = 0; frameIdx < framesList.size(); ++frameIdx) {
    const string& frameName = framesList[frameIdx];
    LOG(INFO) << "getting side camera images from frame: " << frameName;
    for (int i = 0; i < numSideCameras; ++i) {
      const string srcImagePath = FLAGS_root_dir + "/vid/" + framesList[0] +
        "/isp_out/" + camModelArray[i].cameraId + ".png";
      LOG(INFO) << "\t" << camModelArray[i].cameraId << ": " << srcImagePath;
      Mat origImage = imreadExceptionOnFail(srcImagePath, CV_LOAD_IMAGE_COLOR);
      Mat imageUndistorted;
      cvUndistortBicubic(
        origImage, imageUndistorted, intrinsic, distCoeffs, intrinsic);
      sideCamImages[frameIdx].push_back(imageUndistorted);
    }
  }

I think the line

const string srcImagePath = FLAGS_root_dir + "/vid/" + framesList[0] +
        "/isp_out/" + camModelArray[i].cameraId + ".png";

should rather contain framesList[frameIdx], otherwise the images for the first frame are repeatedly loaded?

[fbriggs]

Good catch, you are right. This didn't have any affect most of the time, because if you run the 'rectify' step from the GUI, it only uses one frame. However, if you run TestRingRectification directly, it is possible to pass it more frames, and potentially get a higher accuracy solution. For example, here is a command to run it with frames 0 and 1:

./bin/TestRingRectification
--logbuflevel -1 --stderrthreshold 0 --v 0
--rig_json_file ./res/config/17cmosis_default.json
--src_intrinsic_param_file ./res/config/sunex_intrinsic.xml
--output_transforms_file ~/Desktop/new_rectify_test/rectify.yml
--root_dir ~/Desktop/new_rectify_test
--frames_list 000000,000001
--visualization_dir ~/Desktop/new_rectify_test/rectify_vis

I'll have a diff fixing this soon. Thanks.