Problem with CCTag and SFM Transform Node:

[Satwik Priyadarshi]

We are using the below given meshroom pipeline on 3 systems and are facing some issues.

The 3 systems are:

1. Workstation: GeForce RTX 3070  (Compute capability 8.6); Intel i7-10700k@3.8 Ghz; 32 GB RAM

2. MSI Laptop: GeForce GTX 1650 with Max-Q Design (compute capability 7.5); Intel i7-10750H@2.6 Ghz; 8 GB RAM

3. AWS Server: instance size - g4dn.8xlarge, GPU - 1, vCPUs - 32, Memory - 128 GB, 

We are trying to perform 3d reconstruction from a set of 500 images with AUTO SCALING by using CCTags and SFM Transform node.

The pipeline used is as shown above:

We have encountered following issues:

1. Workstation: On the workstation the performance of SFM Transform node is random. Sometimes cctags are detected and sometimes the CCTag detection fails entirely

2. MSI LAPTOP: On MSI laptop, the CCTags detection and SFM Transform node works perfectly all the time

3. AWS Server: On the AWS server,  the CCTags detection and SFM Transform node fails completely  all the time

It would be great if any one can help us in understating the reason for this random behavior.

Does this depends on hardware, or anything else?

Or is this completely random, due to the deterministic nature of the process?

[Fabien Castan]

Hi Satwik,

Thanks for the precise report.

It seems that you are making the FeatureMatching separately for features and for CCTag marker and that's a good practice to ensure to get a maximum of CCTag markers. In the FeatureMatching with CCTag only, have you properly disabled the geometric validation?

If not, you should do it and redo the comparison.

If yes, then it would be interesting to analyze on the same scene at what point to you start noticing differences?

Are the CCTag correctly detected on all machines? Are all the CCTags correctly matched? Or is the difference during the StructureFromMotion node?

Best,

Fabien

---

De : alice...@googlegroups.com <alice...@googlegroups.com> de la part de Satwik Priyadarshi <sat...@dentratech.com>
Envoyé : lundi 17 janvier 2022 09:54
À : AliceVision <alice...@googlegroups.com>
Objet : Problem with CCTag and SFM Transform Node:

 

** CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe. **

--
You received this message because you are subscribed to the Google Groups "AliceVision" group.
To unsubscribe from this group and stop receiving emails from it, send an email to alicevision...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/alicevision/9f854b55-41a9-4ddb-9dbc-a724183ec9ccn%40googlegroups.com.

[Satwik Priyadarshi]

Hi!

Yes, we are using  two FeatureMatching nodes one for Shift+Akaze and one for  CCTag markers. We have also kept the geometric validation off for the FeatureMatching node  with CCTag markers as suggested by you. 

Moreover, we tried with both “force CPU extraction”  option on and off, but the results were the same.

Actually, we are getting errors in FeatureExtraction node when CCtag3 is on.

Screenshots of pipeline (with error warning) and FeatureExtraction node is attached herewith.

The log file is as follows:

[11:23:14.266476][info] Extracting cctag3 features from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra04026.bmp' [cpu]

[11:23:14.271463][info]       66 akaze features extracted from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra02017.bmp'

[11:23:14.271463][info] Extracting cctag3 features from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra02017.bmp' [cpu]

[11:23:14.295398][info]       604 akaze features extracted from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra05090.bmp'

[11:23:14.295398][info]       178 akaze features extracted from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra01520.bmp'

[11:23:14.297393][info] Extracting cctag3 features from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra01520.bmp' [cpu]

[11:23:14.297393][info] Extracting cctag3 features from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra05090.bmp' [cpu]

[11:23:14.303377][info]       547 akaze features extracted from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra05013.bmp'

[11:23:14.304375][info]       711 akaze features extracted from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra00827.bmp'

[11:23:14.305372][info] Extracting cctag3 features from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra05013.bmp' [cpu]

[11:23:14.308365][info] Extracting cctag3 features from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra00827.bmp' [cpu]

[11:23:14.310358][info]       67 akaze features extracted from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra04425.bmp'

[11:23:14.311355][info] Extracting cctag3 features from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra04425.bmp' [cpu]

[11:23:14.314348][info]       204 akaze features extracted from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra04019.bmp'

[11:23:14.314348][info] Extracting cctag3 features from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra04019.bmp' [cpu]

[11:23:14.351249][info]       1608 akaze features extracted from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra00246.bmp'

[11:23:14.351249][info] Extracting cctag3 features from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra00246.bmp' [cpu]

[11:23:14.382166][info]       360 akaze features extracted from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra05293.bmp'

[11:23:14.382166][info] Extracting cctag3 features from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra05293.bmp' [cpu]

(0) start: 0ms

(1) beforecctagMultiresDetection: 102ms

(2) aftercctagMultiresDetection_inner: 512ms

(3) after marker projection: 0ms

(4) aftercctagMultiresDetection: 2ms

(5) aftercctag::identification::identify: 0ms

(0) start: 0ms

(1) beforecctagMultiresDetection: 99ms

(2) aftercctagMultiresDetection_inner: 514ms

(3) after marker projection: 0ms

(4) aftercctagMultiresDetection: 3ms

(5) aftercctag::identification::identify: 0ms

[11:23:15.566407][info]       0 cctag3 features extracted from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra04026.bmp'

[11:23:15.566407][info]       0 cctag3 features extracted from view 'E:/DB/Scaling Teeth with CCTAG/ankit scan/Lower jaw/i1/dentra02017.bmp'

[Simone Gasparini]

Try using Max Nb Threads = 1 in the FeatureExtraction node when CCTag are used.

S.

To view this discussion on the web visit https://groups.google.com/d/msgid/alicevision/18274cf3-6067-4857-9151-73fe22fee630n%40googlegroups.com.