DepthMap fail since release 2019.1 #409
Labels
bug
for actual bugs (unsure? use type:question)
Comments
|
Can you put the verbose param to "trace" on the DepthMap node and launch the computation again to see if you retrieve more log? |
|
Yep : |
|
I am experiencing a very similar problem on Linux (Ubuntu). I have image sets that reconstructed flawlessly in Meshroom-2018.1.0, but they now fail in the DepthMap node when I run them in Meshroom-2019.1.0. I am including a log trace below. A segmentation fault (core dump) occurs while processing a chunk in the midst of the DepthMap phase. The segmentation fault appears to happen during Log: |
|
I think I have the same problem, here's the error log: It's really sad, given that new version performs depth times twice as fast... |
|
Also worth mentioning is that the error is reproducible every time on the same step, so it's probably bug. Maybe someone is still using C style pointers? |
|
Hello, |
|
Ok, so I have the same issue with 2019.1, also the DepthMap step. I also reran the alicevision command with the trace, and can report that it finishes with status 0. Running from meshroom (50.log, 50.statistics, 50.status): Depthmap_50.zip From terminal with "--verboseLevel trace": alicevision.output_trace.txt It does write the files 832415826_depthMap.exr and 832415826_simMap.exr, but they are smaller than the others: exr-files.zip My config: |
|
Same Here:
[18:10:44.587462][info] CUDA-Enabled GPU. [18:10:44.891280][info] Supported CUDA-Enabled GPU detected. Trace: Program called with the following parameters:
[08:11:04.034628][info] CUDA-Enabled GPU. [08:11:04.299742][info] Supported CUDA-Enabled GPU detected. [08:11:07.053278][debug] [IO] Read Image: C:/Users/renan.goncalves/Desktop/Minis/Druid/MeshroomCache/PrepareDenseScene/ec6132147c1b3835300000fbd60f62a1c989498f/307624374.exr [08:11:09.611046][debug] [IO] Read Image: C:/Users/renan.goncalves/Desktop/Minis/Druid/MeshroomCache/PrepareDenseScene/ec6132147c1b3835300000fbd60f62a1c989498f/1458715582.exr [08:11:12.703855][debug] [IO] Read Image: C:/Users/renan.goncalves/Desktop/Minis/Druid/MeshroomCache/PrepareDenseScene/ec6132147c1b3835300000fbd60f62a1c989498f/899882799.exr [08:11:14.823625][debug] [IO] Read Image: C:/Users/renan.goncalves/Desktop/Minis/Druid/MeshroomCache/PrepareDenseScene/ec6132147c1b3835300000fbd60f62a1c989498f/364451714.exr [08:11:16.948348][debug] [IO] Read Image: C:/Users/renan.goncalves/Desktop/Minis/Druid/MeshroomCache/PrepareDenseScene/ec6132147c1b3835300000fbd60f62a1c989498f/1021059025.exr [08:11:19.228975][info] Estimate depth map, view id: 157664079 [08:11:21.480881][debug] copy image from disk to GPU Elapsed time: 00 minutes 02 seconds 194 miliseconds [08:11:21.480881][debug] rc: 6 |
|
D'oh! I thought we just esablished a habit of not pasting everything into the comment field ;-) |
|
@louisiukas And did any of the devs manage to reproduce it in a breakpoint kind of fashion? |
|
@Darker I have no idea. Just a user and not in the loop. I only commented on the comment trail. |
|
+1 I have the same fail |
|
Same here, just started using meshroom, fails at depthmap. |
|
+3 |
|
Hi, new here and I have the same error, I put it in a pastebin, thanks: |
|
Thank you for all your precise reports. Unfortunately, we never had this problem on all the machines we tested. So I have no idea how to reproduce it for now. |
|
The problem concerns people having problems with the version "2019.1" on the depth map while it worked under the same conditions with the release "2018.1". For people using Meshroom for the first time, you probably just have to update your drivers to be able to use the cuda version used by Meshroom. |
|
Thanks for replying Fabian. I was running the latest windows 7 Nvidia
drivers for my GTX 750ti. I will try to run older drivers. I also have not
tried a different photo set. Will report back.
…On Tue, Mar 12, 2019, 19:38 Fabien Castan ***@***.***> wrote:
The problem concerns people having problems with the version "2019.1" on
the depth map while it worked under the same conditions with the release
"2018.1". For people using Meshroom for the first time, you probably just
have to update your drivers to be able to use the cuda version used by
Meshroom.
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
<#409 (comment)>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/ABykslnd_3zlBRpBlJpL9lFfOjd_YMOTks5vV_Q8gaJpZM4bZ47o>
.
|
|
Hi, |
|
Hi, I have a Gtx 765M and I don't know what is the error, I uploaded the log if somebody wants to take a look, thanks: |
|
I was having this problem with error in DepthMap processing. At first I though was my GPU CUDA disabled, then I read the log saying my GTX105 was detected CUDA enabled. Then I noticed this : [00:18:00.098734][info] Found only 0/10 nearest cameras for view id: 1615417882 I realised it was due to bad photos that the system could not find matching among other photos and stuck here. So I removed the photos with red color icon and to start over the process. Voila it went thru. Hope it helps. |
|
Thanks to @fabiencastan and the creators for making such helpful software readily available. The 2018.1 workflow includes a CameraConnection node. The 2019.1 release removed this. Perhaps this is somehow involved? Also, the log of a successful 2018.1 DepthMap includes many examples similar to There are none in the 2019 log prior to failure. I appreciate the code could have changed. I hope this helps somehow. |
|
@fabiencastan Have a look on rayloo's comment. Maybe you couldn't reproduce the error because you had a good image set without imgaes which cannot be allocated. |
|
Latest update. I had 2 DepthMap failed processes, the 1st one made it after I removed the bad photos. The 2nd did not make it, however the process improved a bit but still stuck before the end of DepthMap procedure. Though it showed 66 over 66 photos recognized in green icon.. |
|
Same here. :-( I just reran image sets in 2019.1 in which all camera positions are properly allocated. Meshroom is still failing in DepthMap. |
|
Following @rayloo this also fails, where we have some cameras I have followed the offending view ids back up stream. The start of the SFM trace is shown here where localizerEstimator has an unknown type. It should be acransac. The SFM html report looks ok. Cheers. |
|
+1 for failure on missing views. Will try deleting the offending views and see if that helps. |
|
Guys, see what I have found. I made 3 projects with DepthMap error. The first one was settled by removing bad photos. The 2nd project did not make it thought I had 100% photos recognized with green icon, but the progress improved a bit. The 3rd project with 100% photos recognized but also failed at about 70% of DepthMap progress. Then I adjusted the Downscale figure to 4 (Default is 2), the result was failed even faster at 10% of the progress. So I thought it was about the resolution issue. Common sense, I adjusted Downscale to 1 and prayed it would go through, and yes it passed DepthMap progress and currently is in the DepthMapFilter progress. Hope my finding means anything to you. |
|
Can confirm that setting downscale to 1 fixes this (at least in my case). |
|
Yep, thank you Fabien ! |
This was referenced Mar 28, 2019
Closed
|
@fabiencastan Any idea when a release would be likely? When it works the multi-GPU support in the 2019 version gave a pretty significant speedup for me (1080Ti x2). |
|
Next release should be around the middle of May. |
Closed
|
Hello,
(I found this was still not working) I have not tested to find if problem 2 is present in release 2018.1.0. I tried it again with the alicevision binaries that come with 2019.1.0 and it still failed so I don't think my compilation is at fault. Admittedly my images weren't the best, but I would expect to just get a bad output rather than fail completely. Here is what I had in aliceVision\bin: bin.zip (irrelevant but when I use those files the 3d viewer refuses to load sfm by double click, no error, very strange) I am using Win 10 x64 and a GTX 1060 Cheers |
|
Unhandled exception at 0x00007FF800CC4398 (aliceVision_depthMap.dll) in aliceVision_depthMapEstimation.exe: 0xC0000005: Access violation reading location 0x0000000000000010. If there is a handler for this exception, the program may be safely continued. |
Closed
|
Are there any additional tips for working around this problem other than lowering the Depth Map Downscale setting to 1 or switching back to the 2018.1 version? |
Closed
|
hitting this pretty consistently when running drone photos of large areas (like a park, building, etc) where there's a couple of rejected photos in the set. any way to get a bugfix release out for 2019.1? or nightly build somewhere? |
|
FYI, downscale to 1 didn't stop it from crashing |
|
What finally fixed this problem for me was realizing that the sensor width for my camera in the cameraSensors.db file is incorrect. The photos I took came from my phones' back camera, but the sensor width stored in the database is for the front/selfie camera. Such a fix surely won't work for everyone, but it's probably worth checking out what sensor width Meshroom is using, anyway. Here's more specific details on finding my sensor width, in case this helps anyone else: |
Any update on the next release timeline? It's nearing the end of June 😕 |
|
@Cdddo the new release is near... Alicevision 2019.2 is nearly complete and for Meshroom a few pull requests need to be merged. |
|
I'm confused. In what scenario does this issue occur? I reconstructed a shoe perfectly fine, but it failed at a banana. What causes this to fail? |
|
You could try the workaround described here: |
|
@fabiencastan Do you think it would be a good idea to just use the 2018 version before this bug? |
|
Isn't 2019.2 supposed to fix this and will be available real-soon? Cheers. |
|
Yes, see #525 for details. |
|
I am having this issue on 2019.2 I have a set of images 1021 and it recognized 1012 images. Might this be a problem? |
|
But should these computations really make the GPU hang / crash, and shouldn't the process report a crash and why instead of just stopping? |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
and others
Hello,
I just checked the new release and I have trouble to render a set of images I've done flawlessly on release 2018.
infobefore stopping.Logs of the chunk that fails :
log
status
{ "status": "ERROR", "execMode": "LOCAL", "nodeName": "DepthMap_1", "nodeType": "DepthMap", "packageName": "aliceVision", "packageVersion": null, "graph": "", "commandLine": "aliceVision_depthMapEstimation --input \"C:/thePath/6966dd1fe97f92b12182da295c478c4c6ae08d3c/sfm.abc\" --imagesFolder \"C:/thePath/479bd69063c5e69424f7008ed63d1d12b05ad0c2\" --downscale 2 --minViewAngle 2.0 --maxViewAngle 70.0 --sgmMaxTCams 10 --sgmWSH 4 --sgmGammaC 5.5 --sgmGammaP 8.0 --refineMaxTCams 6 --refineNSamplesHalf 150 --refineNDepthsToRefine 31 --refineNiters 100 --refineWSH 3 --refineSigma 15 --refineGammaC 15.5 --refineGammaP 8.0 --refineUseTcOrRcPixSize False --exportIntermediateResults False --nbGPUs 0 --verboseLevel info --output \"C:/thePath/eea85f62a67692bf1c80dd467ba452ca1393560d\" --rangeStart 12 --rangeSize 3", "env": null, "startDateTime": "2019-03-01 22:37:08.124094", "endDateTime": "", "elapsedTime": 0, "hostname": "***", "sessionUid": "04b2976e-3c67-11e9-9d66-30243267bd9e", "returnCode": 3221226505, "elapsedTimeStr": "0:00:00" }My config:
Thanks :)
The text was updated successfully, but these errors were encountered: