| Message ID | 20201027141246.4708-1-david.plowman@raspberrypi.com |
|---|---|
| Headers | show |
| Series |
|
| Related | show |
Hi everyone Can I give this patch a little prod, see how folks would like to move forward on this one? Thanks! David On Tue, 27 Oct 2020 at 14:12, David Plowman <david.plowman@raspberrypi.com> wrote: > Hi everyone > > I wanted to raise a topic on which there was already some discussion a > few weeks back, prompted by my noticing a gap in the metadata that we > provide with images. > > My observation was that I couldn't correctly set the ISO value in my > JPEG files because we don't report how much digital gain has been > supplied by the image processing pipeline. To address this problem, it > seemed as good a way as any actually to include a proposal for a new > DigitalGain control (see the associated patch). Nevertheless, I think > there's a discussion to be had first. Notably: > > * Should we let the pipeline report a single global "digital gain" > value or, given that different gains may be applied to the colour > channels, should we report three gain values instead? > > * In the Pi world I'd like this to be a read-only control, i.e. you > can't force the pipeline digital gain to a particular value, as our > AGC doesn't work like that. But there may be platforms that do allow > you to set the digital gain. > > * You could imagine this being "per-stream". I think it's another case > of "practically everyone will just want a single value", though > technically you might be able to imagine some platform and use-cases > where different outputs might reflect different digital gains. > > * There's a question about digital gain being applied by the sensor > itself, but I'm inclined to view that as a separate topic. The issue > at hand, at least for me, is being able to distinguish the gain > applied by the sensor, and which we see in the raw frames, from that > applied by the ISP, which appears in the JPEGs. > > You may have other comments too - so everyone's thoughts would be much > appreciated. > > Thanks! > David > > David Plowman (1): > libcamera: controls: Add DigitalGain control > > src/libcamera/control_ids.yaml | 11 +++++++++++ > 1 file changed, 11 insertions(+) > > -- > 2.20.1 > >
Hi again Can I give this discussion another little metaphorical shove, please? Thanks! Best regards David On Mon, 9 Nov 2020 at 07:53, David Plowman <david.plowman@raspberrypi.com> wrote: > Hi everyone > > Can I give this patch a little prod, see how folks would like to move > forward on this one? > > Thanks! > David > > On Tue, 27 Oct 2020 at 14:12, David Plowman <david.plowman@raspberrypi.com> > wrote: > >> Hi everyone >> >> I wanted to raise a topic on which there was already some discussion a >> few weeks back, prompted by my noticing a gap in the metadata that we >> provide with images. >> >> My observation was that I couldn't correctly set the ISO value in my >> JPEG files because we don't report how much digital gain has been >> supplied by the image processing pipeline. To address this problem, it >> seemed as good a way as any actually to include a proposal for a new >> DigitalGain control (see the associated patch). Nevertheless, I think >> there's a discussion to be had first. Notably: >> >> * Should we let the pipeline report a single global "digital gain" >> value or, given that different gains may be applied to the colour >> channels, should we report three gain values instead? >> >> * In the Pi world I'd like this to be a read-only control, i.e. you >> can't force the pipeline digital gain to a particular value, as our >> AGC doesn't work like that. But there may be platforms that do allow >> you to set the digital gain. >> >> * You could imagine this being "per-stream". I think it's another case >> of "practically everyone will just want a single value", though >> technically you might be able to imagine some platform and use-cases >> where different outputs might reflect different digital gains. >> >> * There's a question about digital gain being applied by the sensor >> itself, but I'm inclined to view that as a separate topic. The issue >> at hand, at least for me, is being able to distinguish the gain >> applied by the sensor, and which we see in the raw frames, from that >> applied by the ISP, which appears in the JPEGs. >> >> You may have other comments too - so everyone's thoughts would be much >> appreciated. >> >> Thanks! >> David >> >> David Plowman (1): >> libcamera: controls: Add DigitalGain control >> >> src/libcamera/control_ids.yaml | 11 +++++++++++ >> 1 file changed, 11 insertions(+) >> >> -- >> 2.20.1 >> >>
Morning all Can I give this one its little weekly nudge? Thanks :) Best regards David On Mon, 16 Nov 2020 at 10:18, David Plowman <david.plowman@raspberrypi.com> wrote: > > Hi again > > Can I give this discussion another little metaphorical shove, please? Thanks! > > Best regards > David > > On Mon, 9 Nov 2020 at 07:53, David Plowman <david.plowman@raspberrypi.com> wrote: >> >> Hi everyone >> >> Can I give this patch a little prod, see how folks would like to move forward on this one? >> >> Thanks! >> David >> >> On Tue, 27 Oct 2020 at 14:12, David Plowman <david.plowman@raspberrypi.com> wrote: >>> >>> Hi everyone >>> >>> I wanted to raise a topic on which there was already some discussion a >>> few weeks back, prompted by my noticing a gap in the metadata that we >>> provide with images. >>> >>> My observation was that I couldn't correctly set the ISO value in my >>> JPEG files because we don't report how much digital gain has been >>> supplied by the image processing pipeline. To address this problem, it >>> seemed as good a way as any actually to include a proposal for a new >>> DigitalGain control (see the associated patch). Nevertheless, I think >>> there's a discussion to be had first. Notably: >>> >>> * Should we let the pipeline report a single global "digital gain" >>> value or, given that different gains may be applied to the colour >>> channels, should we report three gain values instead? >>> >>> * In the Pi world I'd like this to be a read-only control, i.e. you >>> can't force the pipeline digital gain to a particular value, as our >>> AGC doesn't work like that. But there may be platforms that do allow >>> you to set the digital gain. >>> >>> * You could imagine this being "per-stream". I think it's another case >>> of "practically everyone will just want a single value", though >>> technically you might be able to imagine some platform and use-cases >>> where different outputs might reflect different digital gains. >>> >>> * There's a question about digital gain being applied by the sensor >>> itself, but I'm inclined to view that as a separate topic. The issue >>> at hand, at least for me, is being able to distinguish the gain >>> applied by the sensor, and which we see in the raw frames, from that >>> applied by the ISP, which appears in the JPEGs. >>> >>> You may have other comments too - so everyone's thoughts would be much >>> appreciated. >>> >>> Thanks! >>> David >>> >>> David Plowman (1): >>> libcamera: controls: Add DigitalGain control >>> >>> src/libcamera/control_ids.yaml | 11 +++++++++++ >>> 1 file changed, 11 insertions(+) >>> >>> -- >>> 2.20.1 >>>
Hi David, On Mon, Nov 23, 2020 at 07:40:33AM +0000, David Plowman wrote: > Morning all > > Can I give this one its little weekly nudge? Thanks :) Sorry for having you nudgin us for three weeks :( I'll try to provide feedback on the integration with the Control framework. I have less valuable comments on the control semantic itself, so I'll skip that part. > [snip] > >>> > >>> Hi everyone > >>> > >>> I wanted to raise a topic on which there was already some discussion a > >>> few weeks back, prompted by my noticing a gap in the metadata that we > >>> provide with images. > >>> > >>> My observation was that I couldn't correctly set the ISO value in my > >>> JPEG files because we don't report how much digital gain has been > >>> supplied by the image processing pipeline. To address this problem, it > >>> seemed as good a way as any actually to include a proposal for a new > >>> DigitalGain control (see the associated patch). Nevertheless, I think > >>> there's a discussion to be had first. Notably: > >>> > >>> * Should we let the pipeline report a single global "digital gain" > >>> value or, given that different gains may be applied to the colour > >>> channels, should we report three gain values instead? > >>> > >>> * In the Pi world I'd like this to be a read-only control, i.e. you > >>> can't force the pipeline digital gain to a particular value, as our > >>> AGC doesn't work like that. But there may be platforms that do allow > >>> you to set the digital gain. Libcamera does not enforce (yet) controls being read-only. What I mean is that, in example, we don't have a flag in the yaml definition that restrict the control from being written by applications. But metadata are read-only by definition and for your case it is enough not to advertise digital gain in the Camera's ControlInfoMap controlInfo_ to prevent application from setting DigitalGain in a Request. Question is now it it is useful to report a ControlInfoMap of supported metadata and their limits. Anyway that should not block you from reporting DigitalGain like you now report other metadata. > >>> > >>> * You could imagine this being "per-stream". I think it's another case > >>> of "practically everyone will just want a single value", though > >>> technically you might be able to imagine some platform and use-cases > >>> where different outputs might reflect different digital gains. I feel the same way as I felt for ScalerCrop: we should be able to report it per stream, but there's no reason to block the definition of a new control waiting for that feature to eventually land. > >>> > >>> * There's a question about digital gain being applied by the sensor > >>> itself, but I'm inclined to view that as a separate topic. The issue > >>> at hand, at least for me, is being able to distinguish the gain > >>> applied by the sensor, and which we see in the raw frames, from that > >>> applied by the ISP, which appears in the JPEGs. As long as it is made clear in the Control definition, and we don't make it very hard to find a non-conflicting name for the sensor's digital gain I think it's fine. Thanks j > >>> > >>> You may have other comments too - so everyone's thoughts would be much > >>> appreciated. > >>> > >>> Thanks! > >>> David > >>> > >>> David Plowman (1): > >>> libcamera: controls: Add DigitalGain control > >>> > >>> src/libcamera/control_ids.yaml | 11 +++++++++++ > >>> 1 file changed, 11 insertions(+) > >>> > >>> -- > >>> 2.20.1 > >>> > _______________________________________________ > libcamera-devel mailing list > libcamera-devel@lists.libcamera.org > https://lists.libcamera.org/listinfo/libcamera-devel