Issue Regarding Sparse Reconstruction

[paidiakileswar]

Hai ,

Issue is : While trying on my dataset which is 360 Deg images, The output Sparse is not reconstructing well.

Image Specifications:

Camera : Insta360 ONE X2 Dataset : Phoenix Image type : JPEG Image size : 4096 * 2048 Image size in mb : 1.5 to 2 mb

Dataset Details

Dataset Phoenix : 303 Images Sample Dataset image :

Output Reconstruction: (While trying Vocab Tree Matching , Exhaustive Matching without POS.txt=> Output Sparse size formed is 8.2 MB )

Please , Can i know How can we improve this 1) Are there are any specific things to be considered when the running of SphereSfm , Please mention them , 2) Can i know what things did you considered here in the dataset (Like camera side and other )

Thanks ! Hope reply from you ,..

[json87]

You should share the used command lines first. The main difference is the camera parameter setting from the command line listed in this repo README.

[paidiakileswar]

Okay my command are :

1

colmap database_creator --database_path ./colmap/database.db

2

time colmap feature_extractor \ --database_path ./colmap/database.db \ --image_path ./images \ --ImageReader.camera_model SPHERE \ --ImageReader.camera_params "1,2048,1024" \ --ImageReader.single_camera 1 \ --ImageReader.pose_path ./POS.txt

3

time colmap spatial_matcher \ --database_path ./colmap/database.db \ --SiftMatching.max_error 4 \ --SiftMatching.min_num_inliers 50 \ --SpatialMatching.is_gps 0 \ --SpatialMatching.max_distance 50

4

time colmap mapper \ --database_path ./colmap/database.db \ --image_path ./images \ --output_path ./colmap/sparse \ --Mapper.ba_refine_focal_length 0 \ --Mapper.ba_refine_principal_point 0 \ --Mapper.ba_refine_extra_params 0 \ --Mapper.sphere_camera 1 \

These are my parameters !!

Hope this helps you to answer my query !

[json87]

I cannot find the issue from the commands. The most possible reason maybe the overlap degree of recorded images.

If possible, you can share your dataset for test.

[paidiakileswar]

Haii @json87 ,

So Is there a way to tweak parameters at each stage So I get good results for my failed cases ! Like,

Feature Extraction
Feature Matching
Mapper 

Eg : Increasing features or make some changes in some parameters , Will it may enhance my sparse Reconstruction As i am getting very poor sparse reconstruction to my data ! (or) What parameters that tweaked and considered by you are only the best ??

Thanks , Hoping a good reply !

[json87]

For indoor scenes, I think the feature matching step is the key issue since there are not enough matches can be found or there are too many false matches in these scenes.

[paidiakileswar]

Okay @json87

This dataset is actually indoor dataset like a floor in building , So I tried like Spatial,Sequential,Exhaustive too , In all cases the SphereSfm is failing ! I tried to tweak parameters still no improvement !

So 1) Can i know what may be a reason why it may failing ! 2) Can we improve anyway any suggestion , to make improvement 3) or This SphereSfm works good for only outdoor scenes well ,than compared to indoor scenes with more objects ??

Hope these answers will help me !!

[json87]

You could consider using learned features or matching methods instead of SIFT.