I have ordered a R-Pi4 too. So I can test the cam in the meantime.
Processing 12MP images would indeed be hard, and I believe rarely necessary. However, this camera is meant for recording high quality images and videos, and for that Jetson Nano could greatly benefit with its ISP and 4k30-capable h265 encoder. Such camera is currently missing from the portfolio, and actually even finding good quality, high-res USB3.0 ones is hard (at least without spending hundreds of $$$).
For the computer vision-purposed sensor, agreed that we could use sth with lower resolution and possibly monochrome to benefit from good low light performance where picute/video quality is not important.
Concerning video and photo quality provided by sensors like IMX477 with so small photosites, we have to be serious.
Small photosite sensor can’t give very good results. If they could, we would find those sensors in hybrid apn and so on.
Those sensors are everywhere (and therefore not very expensive) because of the smartphone market. Smartphone can give quite good quality because of the hardware and software behind the sensor.
I don’t say IMX477 must be thrown in a trash but i say this is not a good sensor and its resolution is too high for the Jetson Nano.
Everyone will be very pleased with this camera for 5 minutes and when they will realize it can’t be really used, everyone will be much less pleased with it.
A 6Mpx sensor can bring very good pictures and high quality video and considering my experience with the Nano, i think 6Mpx is the maximum resolution a Nano can manage if you plan to make complex treatments with the image or video.
Of course, i can be wrong but as to me, i won’t buy this brand new camera because i know it will be useless. For sure, 50$ is not really expensive but if it does not much your need, then you loose 50$.
I still think the same camera with IMX385 or IMX178 (for higher resolution) would be a must have, even for some extra $.
You’re possibly right that processing at full resolution in real time with any sort of complexity might not be possible, but you can scale down first or bin and scale, and that’s cheap.
I’m aware it’s not the best sensor, and the optics probably aren’t great on a lens that cheap, but it’s still a major upgrade from the Pi camera V2. I’m not expecting it to match my SLR, but it’s certainly an upgrade.
You could very well be right and people hate it. I ordered it as soon as I saw it because I knew it would be out of stock. Maybe I’ll regret it. It has certainly happened before.
For image quality, it depends to what we compare. Larger-pixel sensor would have better quality for the same resolution, and be more expensive. Currently supported IMX219 is only 1/4 at 8 MP though, so 1/2.3 would be a huge step up. And the fact that we find such sensors in smartphones, action cameras and e.g. drones is, in my opinion, precisely due to their decent image quality at low price.
As for nano’s maximum resolutions: object recognition at 12 MP is probably out of question. But ISP can process 1400 MP/s (link), and its encoder 4K video (4K = over 8 MP) at 30fps (link), so 12 MP can easily be utilized for e.g. streaming or recording 4K video. Not to mention photos which are far less demanding.
All in all, IMHO IMX477 support would be a great addition for JN, not for CV or DSLR-class images but for very decent photo and video at affordable price, especially now that thanks to RPi cameras electrically compatible with nano’s connector will be widely available.
@mdegans One of major advantages of RPi HQ camera is the ability to attach any, however expensive, lens with CS mount (or any other via adapter).
maybe am i to severe with this poor IMX477. I do love extreme conditions for photo and video (mainly very low light, poor contrast, heavy noise) and i made some tests with several cameras and sensors, always hoping low price will give good result.
Most of time, it was disappointment. The same disappointment if you want to make Battlefield I work perfectly with a GeForce 256 !
Anyway, we will see if IMX477 can bring interesting things. It will probably bring improvements if we consider RPi Cam V2 (which have a mono version !).
For my personal use, i want something better.
To give you an example of my personal quest (real time video of very low light scene), i have just bought a new camera (Fujifilm XT3 with APS-C Xtrans 4 sensor). Here is a video i made yesterday. It was 10.30pm and the sky was very bad (waxing gibbous Moon, clouds) and the stars were very hard to see (i was able to see 3 or 4 stars with my eyes.
It’s 25fps video (i just made some small adjustments with SkyNano but nothing really important) :
The lens i used with the XT3 is a old Canon FD 50mm F/D 1.4
I think the result is interesting. And yes, a Fuji XT3 is much much expensive than the new RPi HD cam. But to get significant result, we need significant technology.
But i hope new RPi HD cam will be great.
Have a nice day and stay safe.
Thx for the tip. I was looking into those earlier. If you have a good recommendation for Nikon F-mount, I’d appreciate it.
That was my first Nvidia card. I saved up money from building PCs for people. I was very happy with it, especially since the previous card was an ATI Rage Fury MAXX with which I was very disappointed. Before that it was Voodoo 2 and on top of integrated AGP graphics with 4 whopping Mb ram. Whoo. Some memories there.
If you want to use a Nikon lens with the camera, you will easily find a Nikon F to C mount adapter but you will need an extra 5mm spacer i think. That kind of adapter is very common.
My first graphic card used a Nvidia Riva TNT ! Then, i bought a GeForce 256, then a GeForce 2.
After that, i made a deal with the Evil, buying a very good ATI Radeon 9800.
Then, i came back to Nvidia with a 7800 GT.
After that, i made a 2nd deal with the Evil, buying a Radeon HD 6850.
Now, i have a GeForce 1060 in my desktop and also a 1060 in my laptop.
And i will make a break with graphic cards because i have already spent too much money in it !
But i remember very clearly the beginning of GPU. It was really great and really impressive. The WOW effect was really huge.
Now, the WOW effect is not so big. Must be a kind of habit.
Received one of these the other day. As expected, it doesn’t work out of the box.
Has anybody tried with the RR or LI drivers?
Are their drivers openly available? But even if you have them, some modifications may be required, e.g. to the number of MIPI lanes used (RPi HQ cam only has 2 out of 4 routed). Also, not sure if other boards do not have some additional controllers that the driver might talk to instead of directly to sensor.
Just received it today, doesn’t work… Someone else got it to do something? Or is it just nog compatible with the jeston nano
Any idea if support from Nvidia is in the works?
I wouldn’t expect it to work. Drivers have to be written. I ordered mine because I also have a bunch of Pi’s.
The only drivers I know of offhand are those by RidgeRun:
But they’re closed source and extremely expensive. You’ll likely have to wait until somebody in the community writes something. I read on a thread somewhere that support is coming to the libcamera library.
It might be nice if Nvidia started using libcamera as well if possible.
My Rapberry Pi High Quality Camera arrived today. I was hot to use it with my Nano. I was sure (never be sure) that there were drivers.
Can somebody point me to some code that will will work?
You can read about the driver that we are creating here: Raspberry Pi HQ Camera Compatibility
Hey, so if anybody wants a temporary solution for High Quality Camera video to a gstreamer pipeline, you can use a pi to stream to the nano.
On a raspberry pi (a Zero is plenty since this uses it’s video encoder hardware for all the work):
raspivid -l -t 0 -o tcp://0.0.0.0:1234
on the tegra device (or anything that supports gstreamer):
gst-launch-1.0 -v tcpclientsrc host=pi-hostname-here port=1234 ! decodebin ! autovideosink sync=false
If you want the pi to run raspivid every startup, first create a user for the task:
sudo adduser --system --ingroup video --no-create-home --disabled-password --disabled-login cameraman
Then make a file called
/etc/systemd/system/raspivid.service (eg. using
sudo touch) and add these contents:
[Unit] Description=Raspicam On %I After=network.target [Service] ExecStart=/usr/bin/raspivid -l -t 0 -o tcp://0.0.0.0:1234 User=cameraman Restart=on-success [Install] WantedBy=multi-user.target
Then you can
systemctl start raspivid to start the service once in the background or to enable it every startup:
systemctl enable raspivid. Add additional raspivid arguments to the ExecStart line as desired.
note: Only one device can connect to raspivid at a time. Not sure if there is an option to increase the connection limit or not.
A fast question, has this camera auto exposure, auto iris or any way to control the captured light level?
I have it, and it won’t adapt the illumination to a any condition change. I must be loosing something because I cannot imagine a possible use for video.