@jsdoc - let me clear up some confusion you still have:
I did find Github pages indicating that Reolink has / is working on an API (IE so homelink can intercept events etc.) via the OMVIF feed, so they're making some type of effort, so perhaps not quite ready to call it a complete pig of a brand (support actually responds, - This is a totally separate issue and has nothing to do with the issue with Reolink and BI. An API is a completely different command.
Not sure we got to the point of you opining on if it's firmware or hardware that's more lacking in the Reolinks (vs. other brands in similar price ranges). It is a little bit of both. Consumer grade cameras will use lower quality sensors. Think 4k on the 1/3" sensor. That sensor is good for 720P, not 4K. Almost every camera in that price range and below suffer from this.
Further, the consumer grade cameras know that the naive consumer favors a bright static image, so the firmware is written to provide that. That comes at a cost of poor motion at night - the blur city and missing bodies of Reos LOL. The consumer grade cameras may "let" you change parameters, but the camera will override any setting that the user puts in that will darken the image too much.
Also Not completely sure I still quite understand the definition of Keyframe yet (deductively from your last response thought it was a % of the FPS used for AI analysis) but I see it seems to be more of a representation of time to properly process the AI or something more of that nature. But totally get your point and the inference in cause -> effect. Definitely on the bad to unusuable side of the spectrum for Deepstack. Key frame has nothing to do with actually processing time. Key frame or Iframe is how often a complete image is taken. In between they are partial to put together the image. If the FPS and iframes match (say they are both 15), then the KEY is 1.00, which means every second BI is pulling the full frame image to check for motion. Most low end consumer cameras will make the iframe double or triple or quadruple the FPS to lower bandwidth needs. This impacts BI and AI (Deepstack) because in certain fields of view, with a low enough KEY, the object could be completely out of the image before the next full frame is taken and analyzed.
So I am now curious why BI doesn't allow the use of the higher def signal for signal processing - seems like for example GPU's could handle that load if the "average" CPU good enough to run Windows could not and help overcome some of this, but of course I'm still coming at this with minimal understanding of Deepstack or night vision AI etc - This has nothing to do with BI, this is a DeepStack issue. Deepstack is a 3rd party platform and that is how it is set up because it doesn't need all the fine details. It is just detecting outlines for object detection. Using too high quality and the response times will take longer as well.
Since I am sort of ridiculously overpowered on the CPU side for basic DVR purposes, let me ask - what's the better value, a camera with great night vision hardware / generic outfeed and (very) fast CPU to let BI do deepstack / manage the triggers, or a quality brand camera with great AI/sensor combo and disable Deepstack in BI for motion triggers and let camera firmware do the work? Knowing that would allow me to make my consumer choice to "step it up" beyond just dabbling with these cameras and getting serious about fit for purpose decisions. Favor the better camera. A fast CPU with a sucky camera that is a complete motion blur does you zero good. Many here have found for the moment that the camera AI is better, but Deepstack is improving all the time. See the thread below with some of my posts. I have an old 4th gen CPU and am running over 30 cameras and the CPU hums under 15%. Mid teens in the daytime, single digits at night, 45-55% during day with a lot of motion.
Lets say for simplicity we're talking just a basic bullet / stationary cam designed for Driveway use (person / vehicle and perhaps license plate detection etc.) where the quality of detection and night image quality / motion capture accuracy was the main objective. I saw from other postings you were a fan of Dahua's AI, and I'm guessing brands like Lorex or others more pointed toward pro grade security systems would be better choices, but admittedly that's without digging to far into the forums. Once you talk about plate recognition, that is another animal. That needs to be a separate dedicated camera just for reading plates. It cannot be an overview camera and LPR. Keep in mind brands like Lorex are Dahua OEM except usually with less quality sensors and internals.
I guess my objective is to see what this spectrum of choice is within "sane" levels of cost in concert with BI, and a presumption that if you get too far down the price vs performance curve, you get to that point of whether you should just buy the compute hardware with the camera system as an all-in-one and sort of obviate the use case for BI and a PC, etc. Minus the fun of tinkering of course.... For most applications, the Dahua and Hik OEMs ranging from $120ish to $250ish is the price vs performance sweet spot. When I was looking at replacing an existing NVR, once I realized that not all NVRs are created equal (the bandwidth is can process is a huge limiting factor), and once I priced out a good one, it was cheaper to buy a refurbished computer than an NVR.
Many of us buy refurbished computers that are business class computers that have come off lease. The one I bought I kid you not I could not tell that it was a refurbished unit - not a speck of dust or dents or scratches on it. It appeared to me like everything was replaced and I would assume just the motherboard with the intel processor is what was from the original unit. I went with the lowest end processor on the
WIKI list as it was the cheapest and it runs my system fine. Could probably get going for $200 or so. A real NVR will cost more than that.
A member here a couple months ago found a refurbished 4th generation for less than $150USD that came with Win10 PRO, 16GB RAM, and a 1TB drive. You won't find a capable NVR cheaper than that...
Blue Iris has a demo, so try it out on an existing computer and see if you like it.
There is a big
Blue Iris or NVR debate here LOL. Some people love
Blue Iris and think NVRs are clunky and hard to use and others think
Blue Iris is clunky and hard to use. I have done both and prefer
Blue Iris. As with everything YMMV...
And you can disable Windows updates and set up the computer to automatically restart in a power failure, and then you have a more powerful NVR with a nice mobile viewing interface.
Blue Iris is great and works with probably more camera brands than most VMS programs, but there are brands that don't work well or not at all - Rings, Arlos, Nest, Some Zmodo cams use proprietary systems and cannot be used with
Blue Iris, and for a lot of people Reolink doesn't work well either. But we would recommend staying away from those brands even if you go the NVR route with one of those brands...
It has come to my attention that certain Dahua cameras come with built in AI that distinguishes between human & cars. I was unaware of this. Well, always saw the AI word used in the reviews but never really thought about it because I thought it was a Dahua NVR integration only. I am currently...
ipcamtalk.com
, hopefully they'll change that in a future update. But that's what (of course) immediately solved it for me and BI's IP auto-detect worked and made it one click setup basically.
. I don't think I'm even recording a sub stream on this new cam, but I'll see what I can do configure it to produce some data for you. I will say the 823A's infared range is no joke, my shed >100 feet from house is lit up bright and with optical zoom it does impress. 
