Align cameras gets different results from standalone Python module than in App

[irtyeru]

Hello,
This may be related to this question, but I wasn't sure. The problem may just be that I'm fairly new to metashape in python, and I hope that the solution to this problem is obvious to someone who isn't me.

I have a model that was made in two parts on a turntable. I've generated the masks for each photo in photoshop as TIFs and saved them off in a folder called "masks". I load all the pictures into one chunk, load the masks, and  I then run the "align photos" workflow step with the settings as shown in the screenshot attached. In short:

• Generic Preselection is checked
• Reset current alignment is NOT checked
• Reference preselection is Source (default)
• Accuracy is High
• Tiepoint limit is 0
• Masks applied to Key points
• Exclude Stationary Tiepoints is checked
• Guided matching is not checked
• Adaptive fitting is checked.

All my photos align, and I get a sparse cloud with a lot of points.

However, when I run the following code, which should do the same thing (I think), only about 4/100 pics align


doc = Metashape.Document()
        if os.path.exists(newproject):
            doc.open(path=newproject)
        doc.save(path=newproject)
           
        #add  a new chunk
        chunk = None
        if len(doc.chunks)==0:
            chunk = doc.addChunk()
        else:
            chunk = doc.chunks[0]
        #add the photos in the specified directory to that chunk.
        images = os.listdir(photodir)
        for i in images:
            if os.path.splitext(i)[1] in [".jpg", ".tiff",".tif"]:
                chunk.addPhotos(os.path.join(photodir,i))
        doc.save()
        #Add Masks if needed.
        maskKeypoints=False
        if config["mask_path"]:
            mp = os.path.join(config["mask_path"])
            ext = config["mask_ext"]
            masktemplate = f"{mp}{os.sep}{{filename}}.{ext}"
            chunk.generateMasks(masktemplate,Metashape.MaskingMode.MaskingModeFile)
            maskKeypoints=True
            doc.save()
        #match photos/align photos
        chunk.matchPhotos(downscale=consts.PhotogrammetryConsts.AlignDownscale["HIGH"],# =1
                           filter_stationary_points=True,
                           reset_matches=True,
                           filter_mask = maskKeypoints,
                           generic_preselection=True,
                           keypoint_limit=40000,
                           tiepoint_limit=0,
                           guided_matching=True)
        chunk.alignCameras()
        doc.save()

Any advice as to why the two methods don't produce the same results is appreciated. I've opened the doc that is created, and all the masks have loaded correctly, and all the pictures are present.

[smiller]

Add subdivide tasks.

chunk.alignCameras(subdivide_task = True)

[Alexey Pasumansky]

Hello irtyeru,

In the code you are using  guided_matching=True, but on the screenshot Guided Image Matching option is disabled. If setting guided_matching to False does not resolve the problem, please share logs for matching+alignment for both runs (via API and GUI).