Forum MicMac

This forum is dedicated the the community of MicMac users


All times are UTC + 1 hour



Post new topic Reply to topic  [ 9 posts ] 
Author Message
Offline

Joined: Oct 2020
Posts: 3
Gender: None specified
PostPost subject: issus with C3DC
Posted: 28 Oct 2020, 20:53 

Hi,
I am trying to 3D reconstruct with CGI images generated from Blender.
The sparse cloud is working fine. But when I run the command (mm3d C3DC BigMac ".*.jpg" Fontaine) to compute the dense cloud, an error occurred, the message is (Bad Value in eToString for enum value).
I don't have any idea what going on.

I uploaded the images, log output file, and the sparse cloud in the shared folder.
https://drive.google.com/drive/folders/ ... sp=sharing

Any help would be appreciated.


Top
  Profile 
 
Offline

Joined: Feb 2019
Posts: 31
Location: Greece
Gender: None specified
PostPost subject: Re: issus with C3DC
Posted: 01 Nov 2020, 13:58 

From the log file, I'd guess that the issue might not be with C3DC, but with a previous step on your workflow.

Saw this line (167 on log file)
Code:
Residual = -nan(ind)

When you get this in Tapioca or Campari execution, means there is a serious error, but log might not show a failure.


Looks like there might be problematic matching and/or orientation/calibration. Sparse cloud noise points also hints at possible mismatching for some of the images.
Try Schnaps after extracting tie points, to identify possible bad images
Code:
mm3d Schnaps ".*jpg"


If there are bad images identified by Schnaps, try to exclude them to see if you get a valid Residual value & results.

It would be helpful if you shared also previous commands used before C3DC, there could also be an issue with relative orientation.


Top
  Profile 
 
Offline

Joined: Oct 2020
Posts: 3
Gender: None specified
PostPost subject: Re: issus with C3DC
Posted: 02 Nov 2020, 15:13 

Hi, ValkanSot,
Thanks for your response.

I run the following command after both Tapioca and Campari, the output is "0 pictures rejected", and the file "Schnaps_poubelle.txt" is empty. I think there are no suspicious pictures, I paste a part of the console output below.
Code:
mm3d Schnaps ".*jpg"


Code:
All sizes:
  * [800,450] => [43,24] windows of [19,19] pixels
Read packs of homol points:
 Picture MONKEY0001.JPG: Found 22 homol files and 4561 raw homol couples.
 Picture MONKEY0002.JPG: Found 21 homol files and 4119 raw homol couples.
 Picture MONKEY0003.JPG: Found 21 homol files and 3612 raw homol couples.
 Picture MONKEY0004.JPG: Found 20 homol files and 3175 raw homol couples.
 Picture MONKEY0005.JPG: Found 26 homol files and 3023 raw homol couples.
 Picture MONKEY0006.JPG: Found 25 homol files and 2911 raw homol couples.
 Picture MONKEY0007.JPG: Found 28 homol files and 2923 raw homol couples.
 Picture MONKEY0008.JPG: Found 27 homol files and 2613 raw homol couples.
 Picture MONKEY0009.JPG: Found 14 homol files and 2309 raw homol couples.
 Picture MONKEY0010.JPG: Found 16 homol files and 2304 raw homol couples.
 Picture MONKEY0011.JPG: Found 19 homol files and 1999 raw homol couples.
 Picture MONKEY0012.JPG: Found 15 homol files and 1025 raw homol couples.
 Picture MONKEY0013.JPG: Found 9 homol files and 325 raw homol couples.
 Picture MONKEY0014.JPG: Found 18 homol files and 495 raw homol couples.
 Picture MONKEY0015.JPG: Found 22 homol files and 512 raw homol couples.
 Picture MONKEY0016.JPG: Found 15 homol files and 483 raw homol couples.
 Picture MONKEY0017.JPG: Found 14 homol files and 396 raw homol couples.
 Picture MONKEY0018.JPG: Found 21 homol files and 534 raw homol couples.
 Picture MONKEY0019.JPG: Found 30 homol files and 818 raw homol couples.
 Picture MONKEY0020.JPG: Found 30 homol files and 1054 raw homol couples.
 Picture MONKEY0021.JPG: Found 29 homol files and 1838 raw homol couples.
 Picture MONKEY0022.JPG: Found 29 homol files and 2401 raw homol couples.
 Picture MONKEY0023.JPG: Found 23 homol files and 3099 raw homol couples.
 Picture MONKEY0024.JPG: Found 22 homol files and 3870 raw homol couples.
 Picture MONKEY0025.JPG: Found 22 homol files and 4561 raw homol couples.
 Picture MONKEY0026.JPG: Found 22 homol files and 4402 raw homol couples.
 Picture MONKEY0027.JPG: Found 21 homol files and 4013 raw homol couples.
 Picture MONKEY0028.JPG: Found 21 homol files and 3494 raw homol couples.
 Picture MONKEY0029.JPG: Found 21 homol files and 3125 raw homol couples.
 Picture MONKEY0030.JPG: Found 26 homol files and 2982 raw homol couples.
 Picture MONKEY0031.JPG: Found 28 homol files and 2809 raw homol couples.
 Picture MONKEY0032.JPG: Found 31 homol files and 2747 raw homol couples.
 Picture MONKEY0033.JPG: Found 24 homol files and 2513 raw homol couples.
 Picture MONKEY0034.JPG: Found 22 homol files and 2537 raw homol couples.
 Picture MONKEY0035.JPG: Found 24 homol files and 2435 raw homol couples.
 Picture MONKEY0036.JPG: Found 21 homol files and 1818 raw homol couples.
 Picture MONKEY0037.JPG: Found 22 homol files and 1001 raw homol couples.
 Picture MONKEY0038.JPG: Found 22 homol files and 568 raw homol couples.
 Picture MONKEY0039.JPG: Found 17 homol files and 468 raw homol couples.
 Picture MONKEY0040.JPG: Found 16 homol files and 455 raw homol couples.
 Picture MONKEY0041.JPG: Found 16 homol files and 490 raw homol couples.
 Picture MONKEY0042.JPG: Found 15 homol files and 432 raw homol couples.
 Picture MONKEY0043.JPG: Found 21 homol files and 545 raw homol couples.
 Picture MONKEY0044.JPG: Found 22 homol files and 844 raw homol couples.
 Picture MONKEY0045.JPG: Found 21 homol files and 1090 raw homol couples.
 Picture MONKEY0046.JPG: Found 19 homol files and 1256 raw homol couples.
 Picture MONKEY0047.JPG: Found 18 homol files and 1358 raw homol couples.
 Picture MONKEY0048.JPG: Found 20 homol files and 1485 raw homol couples.
 Picture MONKEY0049.JPG: Found 18 homol files and 1393 raw homol couples.
 Picture MONKEY0050.JPG: Found 20 homol files and 1301 raw homol couples.
Found 8734 Homol points (incl. 806 bad ones): 9% bad!
Create new homol....................................................Done!
Write new Packs:
 - MONKEY0001.JPG: 70.6395% of the picture covered (501 points)
 - MONKEY0002.JPG: 68.7016% of the picture covered (472 points)
 - MONKEY0003.JPG: 68.7016% of the picture covered (472 points)
 - MONKEY0004.JPG: 67.2481% of the picture covered (456 points)
 - MONKEY0005.JPG: 70.8333% of the picture covered (511 points)
 - MONKEY0006.JPG: 71.8992% of the picture covered (540 points)
 - MONKEY0007.JPG: 75% of the picture covered (568 points)
 - MONKEY0008.JPG: 79.0698% of the picture covered (648 points)
 - MONKEY0009.JPG: 82.0736% of the picture covered (704 points)
 - MONKEY0010.JPG: 76.938% of the picture covered (595 points)
 - MONKEY0011.JPG: 68.7984% of the picture covered (474 points)
 - MONKEY0012.JPG: 60.1744% of the picture covered (383 points)
 - MONKEY0013.JPG: 37.2093% of the picture covered (177 points)
 - MONKEY0014.JPG: 49.3217% of the picture covered (245 points)
 - MONKEY0015.JPG: 45.7364% of the picture covered (215 points)
 - MONKEY0016.JPG: 44.0891% of the picture covered (228 points)
 - MONKEY0017.JPG: 43.6047% of the picture covered (237 points)
 - MONKEY0018.JPG: 49.6124% of the picture covered (263 points)
 - MONKEY0019.JPG: 46.8023% of the picture covered (253 points)
 - MONKEY0020.JPG: 49.031% of the picture covered (262 points)
 - MONKEY0021.JPG: 56.0078% of the picture covered (331 points)
 - MONKEY0022.JPG: 58.9147% of the picture covered (353 points)
 - MONKEY0023.JPG: 62.9845% of the picture covered (398 points)
 - MONKEY0024.JPG: 67.4419% of the picture covered (457 points)
 - MONKEY0025.JPG: 70.6395% of the picture covered (501 points)
 - MONKEY0026.JPG: 70.6395% of the picture covered (499 points)
 - MONKEY0027.JPG: 68.7016% of the picture covered (474 points)
 - MONKEY0028.JPG: 67.5388% of the picture covered (457 points)
 - MONKEY0029.JPG: 66.7636% of the picture covered (449 points)
 - MONKEY0030.JPG: 69.7674% of the picture covered (489 points)
 - MONKEY0031.JPG: 70.8333% of the picture covered (503 points)
 - MONKEY0032.JPG: 74.031% of the picture covered (551 points)
 - MONKEY0033.JPG: 74.031% of the picture covered (549 points)
 - MONKEY0034.JPG: 78.1008% of the picture covered (611 points)
 - MONKEY0035.JPG: 73.7403% of the picture covered (544 points)
 - MONKEY0036.JPG: 67.7326% of the picture covered (463 points)
 - MONKEY0037.JPG: 59.4961% of the picture covered (360 points)
 - MONKEY0038.JPG: 50.5814% of the picture covered (259 points)
 - MONKEY0039.JPG: 49.1279% of the picture covered (250 points)
 - MONKEY0040.JPG: 54.6512% of the picture covered (299 points)
 - MONKEY0041.JPG: 54.5543% of the picture covered (301 points)
 - MONKEY0042.JPG: 52.907% of the picture covered (303 points)
 - MONKEY0043.JPG: 53.1008% of the picture covered (305 points)
 - MONKEY0044.JPG: 61.531% of the picture covered (379 points)
 - MONKEY0045.JPG: 64.3411% of the picture covered (415 points)
 - MONKEY0046.JPG: 66.0853% of the picture covered (440 points)
 - MONKEY0047.JPG: 64.5349% of the picture covered (417 points)
 - MONKEY0048.JPG: 64.3411% of the picture covered (414 points)
 - MONKEY0049.JPG: 65.407% of the picture covered (429 points)
 - MONKEY0050.JPG: 62.8876% of the picture covered (396 points)
0 pictures rejected.


the workflow command  is below:
Code:
Tapioca MulScale .*.JPG 300 -1 ExpTxt=0 NbMinPt=5

Tapas RadialStd .*.JPG Out=Arbitrary SauvAutom=NONE ExpTxt=0

Campari .*.JPG Arbitrary campari_ori

AperiCloud .*.JPG campari_ori Out=AperiCloud.ply ExpTxt=0

C3DC BigMac ".*.JPG" campari_ori


Top
  Profile 
 
Offline

Joined: Mar 2013
Posts: 238
Location: UMR MAP (3495CNRS/MCC)
Gender: None specified
PostPost subject: Re: issus with C3DC
Posted: 02 Nov 2020, 17:19 

Hello,

Despite the fact that image source are not ideal, I confirm that indeed the main problem is not C3DC but small issues from past commands (actually from the dataset itself) if not the entire workflow.

C3DC seems to fail, because of bad/wrong orientations, because of bad/wrong/not enough tie-points, because of bad dataset.
In short there is almost nothing done properly...
Below some question/optimization point by point.

1/ Tapioca MulScale .*.JPG 300 -1 ExpTxt=0 NbMinPt=5 :
- 300px is kind small to benefit from two pass of MulScale
- Even with "perfect" rendered images, I'm not sure that -1 really helps
- Probably as you can control the camera position and sequence, other mode might be more relevant (Line, File ?)

2/ Tapas RadialStd .*.JPG Out=Arbitrary SauvAutom=NONE ExpTxt=0
- I would use even simplier mode "RadialBasic" as you try to solve "fake photograph"
- why not providing direclty camera positions ?
- Adding RefineAll=0 to be less constrained could help
- You should provides and freeze zero distorsion coef
- An image is not even aligned (consider removing it)
- Try to interpret the log of Tapas/campari : RES:[monkey0048.jpg][C] ER2 1.00884 Nn 76.8437 Of 678 Mul 437 Mul-NN 318 Time 0.03 means that only 76% of tie-points are usable and only 318 used (while it should be AT LEAST 1K points per pictures, better around 3-5K for real images at least)

3/ Campari .*.JPG Arbitrary campari_ori :
- Campari is not necessary as long you don't provide GCP's

4/ AperiCloud .*.JPG campari_ori Out=AperiCloud.ply ExpTxt=0 :
- You should have probably noticed that your cloud is very sparse (because of too few matched features)
- You should have noticed that the scene is inconsistent (cameras misaligned-mismatched, fake convergence somewhere, spilt component?)

5/ C3DC BigMac ".*.JPG" campari_ori :
- this is not the main issue, but be consistent if you wanna use ascii features, use it all along the workflow (add ExpTxt)

On top of that, you try to trick photogrammetric process by providing sub-optimal dataset you can improve by :
- increase image resolution
- using random (and NOT REPETITIVE) and SHARP pattern
- avoid uniform color on the object AND the background
- Better frame the object
- Generate way more camera position as you are in a simulated space !

I would not question the experiment itself, but maybe giving us a why could help us to help you better ;)


Top
  Profile 
 
Offline

Joined: Feb 2019
Posts: 31
Location: Greece
Gender: None specified
PostPost subject: Re: issus with C3DC
Posted: 02 Nov 2020, 17:23 

Schnaps shows that input and tie matching is OK.


Try cropping your sparse model before running C3DC (to avoid large outliers):

Code:
mm3d SaisieMasqQT "AperiCloud.ply"


then add the masked point cloud selection in C3DC command:

Code:
mm3d C3DC [...] Masq3D="AperiCloud_polyg3d.xml" [...]


Top
  Profile 
 
Offline

Joined: Mar 2013
Posts: 238
Location: UMR MAP (3495CNRS/MCC)
Gender: None specified
PostPost subject: Re: issus with C3DC
Posted: 04 Nov 2020, 09:04 

@ValkanShot, it could be helpful (too clean the mess) but it will not solve the fact that the orientation is bad. Plus, you can filter outlier directly in Apericloud using [Name=SeuilEc] REAL :: {Max residual (Def=10)}


Top
  Profile 
 
Offline

Joined: Oct 2020
Posts: 3
Gender: None specified
PostPost subject: Re: issus with C3DC
Posted: 05 Nov 2020, 00:23 

Hi, thanks for your response,
At first, I tried to add a mask on the sparse cloud using "SaisieMasqQT", but the same error occurred in C3DC.

I tried to fix the model, increase the resolution to 3072*1620, move the useless background and object take up a greater proportion of the image, and generate 80 images.

But the higher resolution does not even give me a sparse cloud but gives me a series of errors.

I ran those command, "All ", "MulScale "

Code:
Tapioca MulScale .*.JPG 1000 1620 ExpTxt=0 NbMinPt=5
Tapioca All.*.JPG 1620 ExpTxt=0 NbMinPt=5


Then I tried "RadialBasic ", "RadialStd ", "Fraser"

Code:
Tapas RadialBasic .*.jpg  Out=Arbitrary  ExpTxt=0  RefineAll=0
Tapas RadialStd .*.JPG Out=Arbitrary  ExpTxt=0  RefineAll=0
Tapas Fraser.*.JPG Out=Arbitrary  ExpTxt=0  RefineAll=0


Those commands' combination gives me the same error "Radiale distorsion abnormaly high"
I saw Anthony's suggestion, apologies, I didn't find a solution to provide camera positions and distorsion coefficients. I'd appreciated it if you can point out to me.
The thing is strange here, the previous images' resolution is pretty low, but the higher resolution doesn't give better tie points matched running "Schnaps". Most of them have 200-700 points, and no image is rejected.

Then I tried "Tapioca Line" mode,
Code:
Tapioca Line .*.JPG -1 3 ExpTxt=0
Tapas RadialBasic .*.jpg  Out=Arbitrary  ExpTxt=0  RefineAll=0
AperiCloud .*.jpg Arbitrary Out=AperiCloud.ply ExpTxt=0
C3DC BigMac ".*.JPG" Arbitrary ExpTxt=0


In this mode, the sparse cloud is quite bad. A dense cloud is generated, but it looks horrible, a lot of information is missed. These two clouds are uploaded into the google drive, the link is below:

The newest images are also uploaded, https://drive.google.com/drive/folders/ ... oaUJJs8S7Y?

Any help will be appreciated. :-D


Top
  Profile 
 
Offline

Joined: Jul 2011
Posts: 1050
Gender: Male
Age: 31
PostPost subject: Re: issus with C3DC
Posted: 05 Nov 2020, 15:27 

Hei,

I told you before, on Reddit I think, that your images have very little texture. All the letters and lines are not really picked up by SIFT, as nothing differentiates a part of a line to another part, or a line crossing with another, or the corner of a letter to another, or part of the flat colors to another. You need unique, complexe, dense textures with some granularity. Try putting some high resolution images of tree bark as textures for instance.

Cheers,
Luc

_________________
Join the MicMac community on Reddit : /r/MicMac/
Don't forget to check the wiki : http://micmac.ensg.eu


Top
  Profile 
 
Offline

Joined: Jul 2011
Posts: 1050
Gender: Male
Age: 31
PostPost subject: Re: issus with C3DC
Posted: 05 Nov 2020, 16:34 

I had a little fun, following my own advice, and the result is pretty great. Would be better if I spend more time making the textures prettier and making sure the lights are good, but anyways...

https://drive.google.com/drive/folders/ ... sp=sharing

_________________
Join the MicMac community on Reddit : /r/MicMac/
Don't forget to check the wiki : http://micmac.ensg.eu


Top
  Profile 
 

Who is online

Users browsing this forum: No registered users and 0 guests

Permissions of this forum:

You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
Post new topic Reply to topic  [ 9 posts ] 


cron
Créer un forum | © phpBB | Entraide & support | Forum gratuit