Suggestions

[232344235342]

Long time fan of the project. A couple of ideas I hope help:

1. I haven't found the code yet, as to how you are converting the DNG's to movie files, or if you are just rendering out the DNGs in Resolve? But worth to check the compression levels-- a known YT hack is to upload the highest quality movie file, I typically will use ProRes 422 10bit or higher. There's no way YT will keep all that info, but it will reduce the compression level. Uploading 264/265 is a compression nightmare on YT. That seems to be one of the biggest drawbacks currently on your videos.
2. Colors are getting better, but there's a sort of compression or limiting on them. I've done some scans on high-end machines (same trailer) and compared it to yours, and while overall the colors are close, things like the skin seem to go very orange
3. Speaking of colors, I understand this a personal choice, but just from my experience working with film and video for over 15 years, I feel as though the black levels on the videos could be crushed a bit.
4. If you are working with the DNG's directly in Resolve, I would play around with decoding them to linear/p3 d65 and then CST > Arri Log-c 3 > REC 709. See if you get better images out of it
5. Seems like stabiliser is being improved dramatically :)
6. The PI camera seems to be a bottle neck in image performance, if you could add an option to use larger sensors with high quality glass, I think the image output could be improved. Perhaps something like a, Sony IMX542 Sensor via Blackfly cameras.
7. I think you are now using the multi-frame capture with median blend technique to improve noise?
8. Some generic ideas like: HDR (multi exp capture), or monochrome multi-color (rgb) flashing instead of single pass color capture.
9. Probably not relevant, but worth to double check compression and bit depth on the capture and conversion. Some Py libs will default to use compression, unless specified.

Not sure if there's anything useful there, but hopefully at least one thing. Thanks for all your hard work!!

[meantux]

1- Yes I use DaVinci Resolve to handle the DNGs, it is amazing how fast and efficiently DaVinci Resolve handles them directly. Be sure to set the (Colors Page)->(lower left "Camera Raw" Tab)->DecodeUsing to Clip and check the "Hightlight Recovery" to use the full 10 bits of the Raspberry HQ Camera. But I don't use prores, I can't afford the disk space it requires, I use a compromise, the HQX in 10bits, but even these I delete for my archives... I keep only one 8bits mp4 optimized for Youtube and one 10bits mp4 with at least double the bitrate. DaVinci produces one HQX version, from it I got scripts that make my Youtube mp4 version and my archival mp4 version 2- My skills in color adjusting are limited, I still got a lot to learn. I do my best, but when this part consumes too much time I have to stop, I cut corners and pass to the next scan. There's also the fact that I try to find a single set of settings that would do for all the parts of the trailer when often scenes are not equally balanced on the same roll. 3- I'm always afraid of losing details, But are you telling me that I crush too much or should I crush more? 4- will try.. 5- I should document how I stabilize nowadays, but in short, (Camera Raw) lift= -100 shadows= -100 contrast= +100 and set exposure until we barely see the film holes. Then in Fusion we recently learned to use the Planar Tracker (there are tons of DaVinci features that I don't know about). One fusion tool to insert before the planar tracker: Miscellaneous->ChangeDepth and set it to int8 (this prevents the GPU from seeing patterns naked to the eyes but still embedded in the HDR data), after tracking is done, remove the ChangeDepth and restore the exposure, lift, etc... 6 and 8- Yes the Raspberry Pi camera is a bottleneck but it's still the most bang for the buck that you can ever get. I did work for a while using the superbe but expensive IMX-183 (20MP monochrome sensor) and used 3 pictures to make monochrome multi-color (what I personnally call trichromy, search that word on my Youtube channel). But I left that part behind to concentrate on something that my DIY followers need, something affordable.... and nothing beats the Raspberry Pi HQ camera atm.... I want to return to the IMX-183 and trichromy, I need the extra resolution for IMAX films, but I do have a lot of software issues to fix to even get back at the point I was a few years ago. As for the glass, I'm using the best I could afford: https://www.edmundoptics.com/p/25mm-dg-series-fixed-focal-length-lens/12156/ , it's better than the one we recommend for the project (which is https://www.edmundoptics.com/p/175mm-fl-f56-blue-series-ir-cut-m12-lens/50884/ ) ALSO the GugusseGUI software offers bracketing-Jpegs as an option, but it's up to the user to convert the jpegs into HDRs, it used to be my method for a while, but at one point I did intensive comparisons between DNGs and bracketing. DNGs won as bracketing wasn't adding much except a lot of extra work. 7- no, should I? 9- I personnaly don't use the jpegs (at least since I stopped doing bracketing) and I never bothered to set the quality=, I think it defaults to 95 in the Python PIL library, it's true that I should increase it to 98 for some of my DIY gang that are stubbornly using jpegs...

You're concerned a lot by my compression, but you're evaluating through Youtube... check this out instead: https://we.tl/t-q1o4a1rGv0

[232344235342]

1. Ah so I suspect, this might be a factor to do with the compression? I know it's a pain to upload the HQX due to the size, but have you considered it? At the very least do an a/b test of both mp4 vs HQX uploaded to YT, to compare the compression and see if you get any improvements? (comparing on a large 4k tv for optimal testing)
2. This is definitely a big factor for 35mm prints, especially the older photochemically finished movies (pre-DI) are a lot more unbalanced, and need more work.
3. Crush just a tad bit down, very close
4. I had a quick play with the DNGS (thanks for sending that over), ns whats going on might be the way pi process the raw, but un-select 'davici managed' under the tone mapping on both CSTs (color space transforms).

Have you tried Resolves automatic cutter? I use it on 35mm trailers, it auto slices up the trailer into shots, then I apply the main base grade to the timeline, and just make small tweaks shot to shot, usually just exposure.

5. That sounds like a great stabilisation workflow
6. (8) Understandable, at least you've experimented with it. While MegaPix helps, sensor size is certainly a key factor in image quality -- the larger the photo sites, the better noise floor, exposure, lens quality etc. Hopefully RasPI will produce a new camera with a larger sensor, while still being affordable.
7. It's certainly worth looking into, it adds time, but the results can be worth it, especially with difficult prints and smaller cameras. I did some test recently, and found 6 frames is ideal, but you should see some results around 3, the higher the better. As the digital grain pattern is not fixed completely, taking multiple photos of the same frame and then merging all the layers using avg, will blend the layers and reduce the noise, and add a slight increase in sharpness. It works bc the noise moves around but the frame remains the same.
8. I would remove the jpg function altogether to stop them using it :P haha

Given the limitations of the PI camera, the scans look very good and I can see a great level of improvement over the years.

I only did a very quick grade, and found I could apply it pretty much to the whole project, just needed to adjust the exposure up/down. I also applied my small recipe to recover the Highlights without pulling down the exp. Default: Quick grade:

Default: Quick grade: