A number of thoughts / considerations here:

• This feels very special case for the general PYME repo, and almost certainly doesn't belong in processing. Maybe this should be in a plugin, or we should write a more generic metadata modification recipe, or potentially add a metadata module to group different metadata modifications together.

• This is also the wrong way to do it - applyFilter() is deprecated and should not be used. More importantly, the code as written modifies the metadata of the input. It is a pretty important convention that the input is not modified, moreover, it is currently only by accident that modifications to the input data propagate to the output - there is no guarantee that the output metadata is not created and before the call to applyFilter. The correct way of doing it would be to implement completeMetadata(im) (which takes a copy of the output image as a parameter and, by convention executes at the end of module exeution) and make the changes there.

• While using Filter as a base, and implementing completeMetadata() (along with a dummy apply_filter() which simply returns the data un-modified) would work, it would not be ideal from a performance stand-point, as it will force a copy of the data in memory. A more efficient implementation would be something along the lines of the following:

class SetOriginMetadata(ModuleBase):
    inputImage = Input('input')
    outputImage = Output('output')

    def run(self, inputImage):
        # create a new image, using the data from the old image
        # this re-uses the reference to the unmodified image data so we don't need to allocate more memory
        # but creates a new, empty, metadata
        im = ImageStack(inputImage.data, titleStub=self.outputImage)
        
        # copy the metadata entries from the original image
        im.mdh.copyEntriesFrom(inputImage.mdh)
        
        # add any module-specific metadata to the output image
        self.completeMetadata(im)

        # return the new image
        return im

    def completeMetadata(self, im):
        '''Do any required modifications to the output metadata"
        # Set origin etc ...

Actually, on second thoughts, don't implement completeMetadata - just make the changes in run() method on the new image metadata. ie ...

class SetOriginMetadata(ModuleBase):
    inputImage = Input('input')
    outputImage = Output('output')

    def run(self, inputImage):
        # create a new image, using the data from the old image
        # this re-uses the reference to the unmodified image data so we don't need to allocate more memory
        # but creates a new, empty, metadata
        im = ImageStack(inputImage.data, titleStub=self.outputImage)
        
        # copy the metadata entries from the original image (note, this is not technically required, as the .execute()
        # method will automatically copy (merge) the input metadata to the output, but it doesn't hurt to be explicit)
        im.mdh.copyEntriesFrom(inputImage.mdh)
        
        # add any module-specific metadata to the output image
        im.mdh['Origin.x'] = #New origin value
        ....

        # return the new image.
        return im

The rationale here is the completeMetadata() is a bit of a relic whose function has mostly been moved to ModuleBase.execute() and should be retired.

Also note that "Origin.x" etc are in units of nm, not pixels, and would need to be Float()s rather than Int()

Thanks, David! I am also not sure either if this is a long-term solution, but making the origin changable does make it easier to overlay two imaging modalities on different cameras, although the origin is actually an 'effective origin' based on users and instrument instead of 'real metadata origin'. Previously I used im.mdh.Camera.ROIOriginX because Origin.x metadata in OI-DIC images are missing (have not looked into why). But I made the suggested changes and adding Origin.x metadata also did work.