What is wrong with video in BeagleY-AI? I attached very custom camera: global shutter sensor imx296 (RPi brand). Custom ISP is implemented using TI VISS API. Everything is in the TI’s source code. I’m waiting for a better weather to perform tests in field.
You have had better luck than me. I am not wasting my time on Ti’s BSP. If it does not work fairly decently out of the box, too bad for them because we move on. 8 hours = $1400 in labor for a $100 board…
The BB mission statement is very dear to me, however at some point reality must be dealt with in regards to profitability and ability to produce a viable product. Even though other solutions cost more, we are able to get the board up with less headaches.
I still use the BBB, that one is a tried and true workhorse, and within its capabilities its the go-to board.
If you have the camera up and running please help these guys out and do a nice write up so they can add it the docs.
TI`s SDKs are like dog-shit! Their Wi-sun stack breaks from version to version! 
1 Like
I understand you very well. TI approach is seems to be flowing: they give you some basic source code and you are free to do anything you want, just don’t ask. I’ve spent more than a year with couple of TI products and feel quite comfortable with it now. I didn’t have any TI or embedded experience before. Eight hours here is not enough, I agree.
My understanding is -
BeagleBone provides their boards cheap and hope to attract more developers.
New board like this one is actually not ready for any meaningful mission until one year later when there are more developers having provided enough effort on it. 
It is FAR from being cheap. It is actually one of the most expensive boards. True cost is how much time you waste trying to do some of the most trivial tasks. That drives the cost off the chart.
Biggest issue with the beagley is no emmc and no license to run the gpu. I did manage to get some overlays made for a couple of the pi relay boards. That turned into a nightmare working with the Ti bsp.
Pretty sure I had the nvme connected to that board. You can work around the lack emmc by using a docker image on the nvme, way better than SD card. Docker config can be adjusted to point to the NVMe and save and execute containers from that location. SD is not bad if your just loading ram up, but if you need to cache and use temp files or even run a db its just not the optimum.
I can say that in our testing, the time from command to gpio action is about half that of the BBB. Outliers were almost half the time of the BBB. So it does have plenty of upside potential for applications that are using gpio.
Don’t mind helping out, just will not get taken advantage of. Whomever is collecting the money needs to step up and at least spend their money to get basics up and running.
Every thing that touches the internet is like a big infomercial. Its not how this was 30 years ago, now it turns my stomach how so many are playing games and manipulating others. Big tech culture has permeated society deeply.
4 Likes
Yes, when I started to use internet, net buddies were real people and never scammed each other money. I still miss the era of clean internet.
Those days are long gone and will never return. It is truly shocking how scummy big tech and other companys and people have gotten.
I just pointed dns to a fresh droplet with a fresh FQDN, and within less than an hour some threat actors were trying ancient PHP exploits to break the site.
Hello. I understand, but you shared a gitlab with attached files for compiling the TI SDK. Unfortunately, following your build instructions, I was unable to run the image. Most likely I’m doing something wrong. I would really appreciate it if you could share more detailed instructions on how to build and run your modified image. Thank you!
I have found that “Raspberry Pi AI HAT+” works out of box under Pi OS.
Do they use Python Tensorflow or Pytouch?
Hi, please tell me, where your build got derailed. Maybe we can figure it out…
I bought this board as it fulfills my hardware requirements on paper. With the last of pennies left in my pocket.
But as i made imx219 and 415 detectable by this board. I realised that there no basic drivers for isp and hevc compression.
I wanted to make my own ti linux sdk image with required drivers, which can run on beagley with full functionality.
After coming across this post i am confused, that weather i can achieve my goal or not.
Is the isp, and hevc encoder working on your custom linux distribution.
1 Like
Hi. When i tried to run compiled model on the beagle, i got this errors.
do you know what the problem is?
logs.txt (8,8 КБ)
Also, when i compiled firmware for c7x, I didnt have the folder ‘vision_apps_eaik’. I had only ‘vision_apps_evm’
I noticed too, but the “vision_apps_evm” folder content is the same as “vision_apps_eaik” in the image.
and what about errors in logs? maybe you know what the problem is?
This could be the reason of that error:
TIDL_RT_OVX : TIDL required buffer size - 200704, buffer size allocated by application 150528
It requires more buffer space.
_evm is for TI’s EVM board, not suitable for EdgeAI. EdgeAI works on _eaik firmware.
Need to set the TISDK_IMAGE variable to edgeai in sdk_builder/build_flags.mak in the PSDK.
1 Like
Hello, I followed your steps: BBY-AI / EdgeAI SDK Changes · GitLab changed the corresponding content and successfully compiled and generated the corresponding uboot and rootfs. I replaced these two parts in my SD boot card (the boot card was burned with tisdk-edgeai-image-j722s-evm.wic.xz). Then I powered on and started, but the serial port did not output (using USB-C, the power supply port is also a serial port), and the indicator light was red. I don’t know which step went wrong
You will need a special UART adapter in order to connect to the serial console:
https://docs.beagle.cc/boards/beagley/ai/02-quick-start.html#beagley-ai-headless
USB-C can work as a network interface, not serial, in some cases. For sure it will not work at the initial boot phase.