cavalli1234 / adalam Goto Github PK

Hello @cavalli1234, thank you for the great work.
I am trying the provided code, brilliant btw, but it looks like the matching could match to the same point right?

The following list is the matches generated from my test.

# testing
adamatcher = AdalamFilter()
adamatcher.config['scale_rate_threshold'] = None
adamatcher.config['orientation_difference_threshold'] = None
adamatcher.config['area_ratio'] = 10
adamatcher.config['search_expansion'] = 10

matches = adamatcher.match_and_filter(kpA, kpB, descA, descB, im1shape=None, im2shape=None, o1=None, o2=None, s1=None, s2=None).cpu().numpy()
 
# matches
[ 19 359]
 [ 23 429]
 [ 24 413]
 [ 53 389]
 [ 54 382]
 [ 55 385]
 [ 74 503]
 [ 84 403]
 [ 95 465]
 [ 99 526]
 [115 543] <---
 [118 490]
 [119 543] <---
 [124 543] <---
 [137 465]
 [147 420]
 [152 589]
 [153 495]
 [172 589]

Is this an expected feature? Thank you

we have a mac, how to setup env for cpu? � Any help would be appreciated.

How do I create a colmap dataset with custom dataset?

• Thank you very much for your article and code. You have provided me a great help on outlier filtering. I have read every line of your code and thesis thoughts. Both the idea and the code are extremely good I have seen,but I have a little doubt about the function of confidence_based_inlier_selection in ransac.py.
• The question is about Code on line 43 in ransac.py( _, inv_indices, res_dup_counts = torch.unique_consecutive(sorted_res_sqr.half().float(), dim=1, return_counts=True, return_inverse=True)
• I guess you want to delete duplicate value in order to prevent this duplicate value to disorder find inliers in uniformly scattered outliers(maybe my English is hard to understand because i am a Chinese...i'm so sorry about it QAQ）for example，after computing of residuals，i got the value of residuals :(0.0000e+00, 0.0000e+00, 0.0000e+00, 1.9729e-05, 7.6246e-04, 1.8478e-02). in order to equip the real residual with construct continuous residuals,i must to delete the value (0.0000e+00, 0.0000e+00, 0.0000e+00),so i can guarantee the rest of values can continuous growth?
• But i find ,in each iteration of 128 iterations，they both user the same index we find duplicate value in first iteration ，but i think each iteration have different values，so their index of duplicate value must be different. Unless the offset of these values in xamples is very small so no matter which A can make its residual very littel?
• I'm so confused about it, so i want to ask for you ,really bother you very much！

When the program example.py is executed, an error is reported, and it seems that the program cannot run successfully.

Hello! Thank you very much for releasing the implementation of this work.
The runtime & performance are quite impressive.

• Here, I would like to ask for your suggestion on how to extend this work. Actually, do you have a version of this work for batch processing over multiple images at the same time? Is it possible?

• Also, could you please recommend how to estimate the scale and orientation components of local features? I have actually used the SuperPoint as the local feature. I am not sure how to get these two components ?

I put the database.db file and the image_pairs_to_match.txt file into the examples folder, and when I run the match_colmap_database_example.py file in the examples folder, using the conmand:

python match_colmap_database_example.py --database_path autodl-tmp/AdaLAM-master/examples --image_pairs_path autodl-tmp/ AdaLAM-master/examples/image_pairs_to_match.txt
which produces an error with the following path:

Traceback (most recent call last):
File "match_colmap_database_example.py", line 11, in
matcher.match_colmap_database(database_path=opt.database_path, image_pairs_path=opt.image_pairs_path)
File "/root/miniconda3/envs/adalam/lib/python3.7/site-packages/adalam/adalam.py", line 189, in match_colmap_database
connection = sqlite3.connect(database_path)
sqlite3.OperationalError: unable to open database file
What is the solution?

hello, I download the code and run example.py
but get empty matches, Is anyone having the same problem?

what about the performance compared with superglue?

Hi @cavalli1234 @Dawars !

Thanks so much for sharing the code. I love reading your paper, and it proves that there are so many good combinations that can create a state of the art performance. I am learning about it from your paper.

Here, I tried to understand how the outlier rejection works from the code. However, I have stumped upon extract_local_patterns Line 60. I wonder what are you trying to do ? Could you please help explain?

Line:60 expanded_local_scores = scores[tokp1] + ransidx.type(scores.dtype)

I wonder why you have added them together?
The score and ransidx are variables that are used for a totally different purpose. The score was containing the ratio of the lowest value and the second-lowest value (I guess it is related to Lowe's thresholding). Meanwhile, the ransidx was from Line 51, which are the row of fnn_to_seed_local_consistency_map_corr whose number of neighbors are higher than a specified value of inliers, or specifically, it is...:

**Line:51**    ransidx, tokp1 = torch.where(fnn_to_seed_local_consistency_map_corr)

Or is this the way you re-order to be the same as ransidx? Could you please kindly explain.... :)

Hi, thank you for your great work.

While testing the code, I find there are some cases where no correspondences survive after filtering. May be ratio test could be adopted in this situation to be a remedy.

Hello,

I'm wondering how the code could be adapted to process 2D-3D correspondences (including descriptors)?

Thank you,
~ Dan

Hello! Thank you very much for releasing the implementation of this work.
The runtime & performance are quite impressive, and it works well for me for the image feature matching.

but how can i modify the parameter listed in the 'DEFAULT_CONFIG' of the 'class AdalamFilter: ', i have tried to modify it, but it does not work for the output.

The default parameter list:

The parameter list after modified:

Hello, why I imported the db file generated by your code into the colmap and it shows that there is no match. I used SuperPoint+SuperGlue for the feature point and the descriptor.

Hi,

Would you be interested in the AdaLAM integration into kornia? We are now developing local features module and AdaLAM would be a great fit.
It would also boost usage and citation of AdaLAM :)

--
Best, Dmytro.

hi @cavalli1234 @Dawars

Thanks for sharing the code, I'm marvelous about the result of the code but also get some questions especially in the function of confidence_based_inlier_selection.

in the formula (4) of the paper,

Could you explain more about the second equation? What does the R stands for? Why Ep can be substituted by the next formula?

And I found it hard to map the formula to the corresponding code as indicated by the pic below.

I supposed it's the place you apply the formula? Or if I'm wrong please tell me.

Thank you in advance