search

alicevision / Meshroom

3D Reconstruction Software
http://alicevision.org
Other
11.07k stars 1.08k forks source link

[bug] Random results #1204

Open refob opened 3 years ago

refob commented 3 years ago

Describe the bug I ran meshroom 6 times using the same setup and got 6 different results (=different looking meshes). I used only one camera (see https://en.openscan.eu/openscan-mini), but meshroom seems to detect many of them at different distances. The same set of images are detected by VisualSFM w/o problems (see screenshots).

To Reproduce Steps to reproduce the behavior:

  1. Add this line "RaspberryPi;RP_imx219;3.68;devicespecifications" to ".../cameraSensors.db"
  2. Start meshroom
  3. Load all images - get them from https://drive.google.com/file/d/1p2av8p83zYrbxBXFK81yXtN3Qe3eHIsz/view?usp=sharing
  4. Run
  5. Look at the result

Expected behavior Reproducible results are expected. Only one camera should be detected at similiar distances from the object.

Screenshots

Model pictures: IMG

Result 2. run: T2

Result 3. run: T3

Result 4. run: T4

VisualSFM camera detection: VisualSFM

Log

 * imageFolder = "" (default)
 * input = "C:\Users\reinhard\AppData\Local\Temp\tmpzr8sprpi/CameraInit/ac38bdd16b58bc6daeecdeacbfc7df059355110f//viewpoints.sfm"
 * output = "C:/Users/reinhard/AppData/Local/Temp/tmpzr8sprpi/CameraInit/ac38bdd16b58bc6daeecdeacbfc7df059355110f/cameraInit.sfm"
 * sensorDatabase = "C:\Tools\Meshroom-2020.1.1\aliceVision\share\aliceVision\cameraSensors.db"
 * verboseLevel = "info"
 * viewIdMethod =  Unknown Type "enum aliceVision::sfmDataIO::EViewIdMethod"
 * viewIdRegex = ".*?(\d+)" (default)

[17:25:21.305270][warning] Some image(s) have no serial number to identify the camera/lens device.
This makes it impossible to correctly group the images by device if you have used multiple identical (same model) camera devices.
The reconstruction will assume that only one device has been used, so if 2 images share the same focal length approximation they will share the same internal camera parameters.
150 image(s) are concerned.
[17:25:21.360214][info] CameraInit report:
        - # views listed: 150
           - # views with an initialized intrinsic listed: 150
           - # views without metadata (with a default intrinsic): 0
        - # intrinsics listed: 1

EXIF data:

Filename - 2020-11-15_19-00-02.jpg
XResolution - 72
YResolution - 72
ResolutionUnit - Inch
DateTime - 2020:11:15 19:00:04
YCbCrPositioning - Centered
ExifOffset - 122
Make - RaspberryPi
Model - RP_imx219
ExifVersion - 0232
DateTimeOriginal - 2020:11:15 19:00:04
DateTimeDigitized - 2020:11:15 19:00:04
ComponentsConfiguration - YCbCr
FocalLength - 3.04 mm
FlashPixVersion - 0100
ColorSpace - sRGB

Desktop (please complete the following and other pertinent information):

Additional context Hardware: Intel i7-5960X CPU, 3.00 GHz, 32GB Ram Windows 10 Pro, Version 2004 GeForce GTX 980 Ti, Nvidia driver 457.30 Cuda V11.1.74

refob commented 3 years ago

Can somebody please take a look at this?

fabiencastan commented 3 years ago

It works fine for me, but I have used the development version: image

refob commented 3 years ago

That's really GREAT news. Is there a compiled Win10 version that I could try?

On 04.06.2021 19:54, Fabien Castan wrote:

It works fine for me, but I have used the development version: image https://user-images.githubusercontent.com/153585/120843539-839b4e80-c56e-11eb-8548-159f12174539.png

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/alicevision/meshroom/issues/1204#issuecomment-854904062, or unsubscribe https://github.com/notifications/unsubscribe-auth/ASKHOTM627OGJRLYLJJLL3DTREHMBANCNFSM4VQMYHNQ.

fabiencastan commented 3 years ago

Unfortunately, we don't have a nightly build system in place... Try to use dspsift instead of sift, it's now the default in the development version.