Hi All , My startup has been using jetson nano for while , its expensive and way too complicated with proprietary NVidia stuff like cuddn / tensorrt /deep stream , how does it compares and can i build custom solution directly with embedding helios chip
Hello @Gaurav_Prasad i have a start up too, after seen that Nvidia Jetson is not a viable solution (difficult to find in the market at reasonable price and constantly) we switched up to Hailo8 with an embedded Linux solution, at the moment raspberry 5 .
You should look at an embedded solution with Yocto and custom hardware, i’m going toward this direction.
Hello @Gaurav_Prasad,
Welcome to the Hailo Community!
Yes, you can use our solutions. We offer the Hailo8 as an accelerator primarily for Linux-based systems (with some Windows compatibility), as well as the Hailo15 SoC which is more oriented toward embedded systems with yocto.
Could you share more details about your intended use case so I can provide you with more specific guidance?
Hey @Andrew92 ,
Happy to hear that the RPI with hailo8 is working for you .
If you’re interested in learning more about our Hailo15 offerings for Yocto and embedded systems, please share additional details about your requirements so we can provide more targeted assistance.
hello @omria i’m interested in learning more about Hailo15 for Yocto since i need to develop a personalized embedded device for my startup, but first i need to test the yolov segmentation with my custom model, as already said in my posts https://community.hailo.ai/t/yolov8-seg-custom-train-config-json-and-posprocess-file-missing/13076/3 https://community.hailo.ai/t/segmentation-fault-on-running-custom-yolov5n-seg-hef/13119 until thi issues are not fixed i cannot move over, because i need to test the segmentation 
if you can help me on this..
thanks @omaria @Andrew92 we are creating AI gate sentry with face rec , LPR , vehicle counting and tracing them inside campus …we are using custom model right now we are using yolo
also wanted to know hardware availability ,is it stable ? my team is really getting tired with Nvidea specially at edge …also is halio is Chinese company or isreally
@Andrew92 which version AI hat are you using 13 or 26 tops ? are you using custom models ? does it only work with UVC camera also or only rpi camera ..what about Hailo-10H M.2 , have you tried it ? looks cool …@omria can we use any x86 or arm with Hailo-10H or 15 m2 ? also do you also have OEM or startup program
Hey @Gaurav_Prasad,
Hailo is an Israeli company specializing in stable hardware solutions for edge AI deployment.
For your needs, we recommend the Hailo-8 rather than the Hailo-10H (which is more oriented toward LLM applications). The Hailo-8 is optimized to work with both x86 and ARM architectures, and we provide examples for most common use cases , Also works with all types of cameras , it based on the app that runs it.
If you want the SOC , then Hailo15.
I’ll have our marketing team reach out to you with more information about our OEM/startup program and specific offerings that might suit your requirements.
1 Like
Hey @Gaurav_Prasad i’m using Hailo8 26tops, talking with @omria , popped out that i could do a guide on how to train, deploy and use custom models.
Unfortunately the problem i have right now is the “Segmentation fault”. Once this problem will be resolved i can share the complete process to manage custom models.
Actually, i’m working with my startup on custom models, i succesfully trained yolov5 (not seg) on my dataset and deployed to raspberry, we use video input from our hardware, practically HDMI to USB input , to recognize objects, but you surely use all type of camera without problem, as an input.
thanks @omria , looking forward to collaborate
1 Like
@Andrew92 …guide seems to excellent idea , i would like to contribute too ..hopefully when we port from nvidia to halios …some quick questions though … whats the fps you are getting Hailo8 26tops , are you using quantized model 8 bit ? how’s the accuracy …and why yolo5 and why yolo 8 or 11 ..
1 Like
hi @Gaurav_Prasad ! so first thing first:
Didn’t see how many fps precisely but our application needs at least 30fps, we are highly above that, it’s incredible because old system run with RTX4060 but we do not see any performance difference or at least, we don’t see any problem caused by different performances, given that we also decide to switch to Hailo also for this reason.
We are using quantized 8 bit models trained in our own datasets, the accuracy is pretty good, but this depends also on what are your targets.
Yolov5 because it comes with GPL3.0 and yet have good perfomance, yolov8 and 11 comes with AGPL and would be way more ocmplicated to integrate in a commercial project, our goal will be to use a yolo with MIT licence.
1 Like