IndexError: _Map_base::at when runing pycolmap.triangulate_points

[bowieshi]

Hi. I am using hloc triangulation to reconstruct my scene when I have camera ground truth pose and images. I first use rec.add_camera(camera), rec.add_image(img), rec.register_image(image_id) to build the rec. I have set img attribute cam_from_world = pycolmap.Rigid3d(pycolmap.Rotation3d(qvec), tvec) with ground truth pose. Then I export them using rec.write(export_dir / 'reference_model'). I got 3 .bin file and I tried using colmap model_converter --output_type TXT to .bin. And I got 3 .txt file. I found my image information looks complete, like following:

# Camera list with one line of data per camera:
#   CAMERA_ID, MODEL, WIDTH, HEIGHT, PARAMS[]
# Number of cameras: 1
0 SIMPLE_PINHOLE 640 480 500.23599999999999 318.72800000000001 240.06800000000001

# Image list with two lines of data per image:
#   IMAGE_ID, QW, QX, QY, QZ, TX, TY, TZ, CAMERA_ID, NAME
#   POINTS2D[] as (X, Y, POINT3D_ID)
# Number of images: 2163, mean observations per image: 0
1108 0.34782986433108698 0.077406669808001469 0.93358963585906773 0.037854785234972349 -2.65252 0.16404199999999999 -0.65826300000000004 0 frame_001108.png

1107 0.33061990066509883 0.080584875788237018 0.94003671756553597 0.022967993099255912 -2.6674899999999999 0.15226100000000001 -0.668041 0 frame_001107.png

1106 0.32005393416087918 0.085187182475924828 0.94354280590137418 0.0059662887726593377 -2.6869499999999999 0.096220100000000003 -0.68079999999999996 0 frame_001106.png

1105 0.30174095045953364 0.091024885055342175 0.94845684428035271 -0.0331131945634058 -2.6903100000000002 0.087598800000000004 -0.70806999999999998 0 frame_001105.png

1104 0.29772088183921863 0.093955062710698853 0.94858562352113918 -0.052153979300897851 -2.6884399999999999 0.070330500000000004 -0.71776899999999999 0 frame_001104.png

1103 0.29673592636449936 0.095477876307010393 0.94715776496126702 -0.07565668122567204 -2.6871 0.047271599999999997 -0.72970100000000004 0 frame_001103.png

1102 0.2921759087227922 0.095757770084796134 0.94689770418444541 -0.094066070613291461 -2.6898900000000001 0.026290299999999999 -0.73790500000000003 0 frame_001102.png

1101 0.29041503705016386 0.093835911971266894 0.94590712067561733 -0.11006201404133788 -2.69685 0.0075736500000000003 -0.74869799999999997 0 frame_001101.png

1100 0.28803302007224268 0.089337106225661489 0.94450006581965695 -0.13029000907955859 -2.6995399999999998 -0.016844600000000001 -0.75598399999999999 0 frame_001100.png

......
......
......

# 3D point list with one line of data per point:
#   POINT3D_ID, X, Y, Z, R, G, B, ERROR, TRACK[] as (IMAGE_ID, POINT2D_IDX)
# Number of points: 0, mean track length: 0

Then, I use the following code to reconstruct my scene under ground truth poses:

def get_reconstruct(images_dir, export_dir):
    # extract global features for retrieval
    feature_conf = extract_features.confs["netvlad"]
    global_features_path = extract_features.main(
        conf=feature_conf, 
        image_dir=images_dir, 
        export_dir=export_dir
    )

    pairs_path = export_dir / 'pairs.txt'
    pairs_from_retrieval.main(
        descriptors=global_features_path, 
        query_prefix="frame",
        db_prefix="frame",
        output=pairs_path,
        num_matched=20,
    )

    # extract local features for reconstruction
    feature_conf = extract_features.confs["superpoint_max"]
    local_features_path = extract_features.main(
        conf=feature_conf, 
        image_dir=images_dir, 
        export_dir=export_dir
    )
    match_conf = match_features.confs["superglue"]
    match_path = match_features.main(
        conf=match_conf,
        pairs=pairs_path, 
        features=feature_conf['output'], 
        features_ref=local_features_path,
        export_dir=export_dir
    )
    reconstruction = triangulation.main(
        sfm_dir=export_dir / 'sfm', 
        reference_model=export_dir / 'reference_model', 
        image_dir=images_dir, 
        pairs=pairs_path, 
        features=local_features_path,
        matches=match_path
    )
    return reconstruction

However, I got strange error when run pycolmap.triangulate_points:

Traceback (most recent call last):
  File "/home/qyzheng/boao/project/scripts/camera_localize.py", line 158, in <module>
    reconstruction = get_reconstruct(reference_images_dir, export_dir)
  File "/home/qyzheng/boao/project/scripts/camera_localize.py", line 96, in get_reconstruct
    reconstruction = triangulation.main(
  File "/home/qyzheng/boao/project/Hierarchical-Localization/hloc/triangulation.py", line 256, in main
    reconstruction = run_triangulation(
  File "/home/qyzheng/boao/project/Hierarchical-Localization/hloc/triangulation.py", line 211, in run_triangulation
    reconstruction = pycolmap.triangulate_points(
IndexError: _Map_base::at

Can I inquire whether I missed something? I looked through all issues but can not find similar problem. I have stuck here for several days. If you can help, many thanks!

[LI-WEN-XUAN]

same problem. did you fix it?

[sarlinpe]

Hi,
Sorry that you are running into troubles. Which version of COLMAP/pycolmap are you using? built from source or downloaded from PyPI?

[LI-WEN-XUAN]

Hi, Sorry that you are running into troubles. Which version of COLMAP/pycolmap are you using? built from source or downloaded from PyPI?

In pipeline.py under Aachen_v1_1, replace feature_conf with “r2d2“ and matcher_conf with “NN-mutual”. The same error is displayed： File "/home/edward_li/vision/vslam/Hierarchical-Localization/hloc/triangulation.py", line 211, in run_triangulation
reconstruction = pycolmap.triangulate_points(
IndexError: _Map_base::at

[FelixMoeller3]

If you experience the same error, it may be worth downgrading your COLMAP installation. The current COLMAP installation is broken for triangulation as can be seen in the given issue

[Penguin-jpg]

Hello, did you solve this problem? I encountered the same problem and cannot fix it even after downgrading the COLMAP.