tsattler / visloc_pseudo_gt_limitations Goto Github PK

Hi,
I would like to know if you can also share the SIFT keypoint descriptor that you used to generate these SFM ground truths, i.e., the db files of the COLMAP. Thanks.

Very impressive work! I notice that the test result of DSAC* on dslam gt in is paper is much much higher than the original DSAC* paper. So, what is the difference of the method between these two version? What may cause the difference?

Hi,
I want to generate poses for query (new) images of a scene using the SFM model built on reference images only.
One way I found is to use image_registrator in COLMAP (not sure if it is the correct way).

But I'm getting below error while running command -

colmap image_registrator     \
--database_path  /data/outputs/hloc/King_Seq1_ref_only/sfm_superpoint+superglue_with_query/database_with_query.db     \
--input_path  /data/outputs/hloc/King_Seq1_ref_only/sfm_superpoint+superglue   \
--output_path  /data/outputs/hloc/King_Seq1_ref_only/sfm_superpoint+superglue_with_query

Std Output-

==============================================================================
Loading database
==============================================================================

Loading cameras... 261 in 0.000s
Loading matches... 33929 in 0.125s
Loading images... 261 in 0.019s (connected 261)
Building correspondence graph... in 2.984s (ignored 682)

Elapsed time: 0.052 [minutes]

F1213 16:50:19.615032 97173 reconstruction.cc:81] Check failed: existing_image.Name() == image.second.Name() (seq1_frame00261.png vs. seq1_frame00211.png)
*** Check failure stack trace: ***
    @     0x7fd54172d0cd  google::LogMessage::Fail()
    @     0x7fd54172ef33  google::LogMessage::SendToLog()
    @     0x7fd54172cc28  google::LogMessage::Flush()
    @     0x7fd54172f999  google::LogMessageFatal::~LogMessageFatal()
    @     0x55b891c61a6d  (unknown)
    @     0x55b891dc8d4e  (unknown)
    @     0x55b891badaf5  (unknown)
    @     0x55b891b9ca0e  (unknown)
    @     0x7fd53d3afb97  __libc_start_main
    @     0x55b891ba66aa  (unknown)
Aborted (core dumped)

The paper On the Limits of Pseudo Ground Truth in Visual Camera Re-localisation has presented that they have used similar technique (without mentioning the COLMAP function) to find ground truth poses for query images. This is the reason I'm posting this issue here.

Am I going the correct way? If not what is the correct method to do so?

Hi @tsattler @ebrach ,
Thanks for your awesome work. I am confused about the coordinate of the .ply pointcloud data from the SfM-based pseudo ground truth (pGT) is OpenGL or OpenCV?

Thank you very much for your work, could you please provide the DenseVLAD image pairs for 12scenes if you have time?

Thank you very much for your work, could you please provide the DenseVLAD image pairs for 12scenes if you have time? I would like to do some experiments and research similar to 7scenes. My contact information is [email protected]

Hi,

I would like to ask what is the implementation of the PnP algorithm that you used to compute the estimated pose of your localization algorithm? I'm asking because I am having some problems with mine (correct 2d-3d matching but large differences w.r.t ground-truth poses). I am currently using openCV's solvePNPRansac. Thanks.

Hi,
Thankyou for your responses to #1 and #2 .
Based on the information provided, I tried to create SFM model and consecutively poses for reference as well as query images. The superpoint and superglue is used for feature extractor and feature matching with exhaustive feature matching.

When I evaluated PixLoc - Localization network, on produced data , the performance is worst (result below).

Cambridge ShopFacade Scene -

[12/17/2021 20:35:54 pixloc INFO] Evaluate scene ShopFacade_ours: /home/ajay/pixloc/outputs/results/pixloc_Cambridge_ShopFacade_ours.txt
[12/17/2021 20:35:54 pixloc.utils.eval INFO]
Median errors: 5.241m, 65.716deg
Percentage of test images localized within:
        1cm, 1deg : 0.00%
        2cm, 2deg : 0.00%
        3cm, 3deg : 0.00%
        5cm, 5deg : 0.00%
        25cm, 2deg : 0.00%
        50cm, 5deg : 0.99%
        500cm, 10deg : 2.97%

Can you please help by sharing full pipeline used for producing ground truth pose data?

#1 Thanks for quick response!

According to reply to #1 , --fix_existing_images is set to True while creating poses for query images.

Passage from Paper

First, we reconstruct the scene with SfM using only the training images. Next, we continue the reconstruction process with the test images while keeping the training
camera poses fixed.

According to this passage, doesn't this mean that --fix_existing_images should be set to False? or am I missing something?