Hi everyone,
I’m new to embedded development and have mainly worked on my computer using Python (VSCode). I’ve previously completed a project detecting and identifying animals in forests using YOLO and ResNet with a bit of fine-tuning on a custom dataset (created from YouTube images)
This time, I want to build an embedded AI module for detecting vehicles in infrared. I’m considering using the NVIDIA Jetson Nano paired with an infrared camera
Is it feasible to train a custom AI model (e.g., YOLO) on my PC, fine-tune it with a custom database, and deploy it effectively on the Jetson Nano for this purpose? I’d like to incorporate my own images or dataset to help the model detect and identify vehicles accurately
Has anyone worked on similar projects or have tips/resources to share?
Any guidance or recommendations would be greatly appreciated
Thanks
Jetson Nano is old and won’t receive any software updates any more. You should not start any new projects on it. Jetson Orin Nano is the current generation but a different product. Same form factor, mostly compatible pinout, much improved performance. Orin NX is the same generation, same form factor, same pinout, but more cores and even more performance, especially for AI.
OK thanks for your answer,
Jetson Nano is maybe outdated and won’t get updates but it’s much cheaper than the Orin Nano and might still handle my task (infrared vehicle detection), dont you think ?
The Raspberry Pi 5 with the Hailo AI Kit could maybe be a cheaper option. The hailo ai kit for rasberry has been released few months ago and is quite powerful, what do you think about that ?
If the software you want to use runs on Ubuntu 18.04 and on a 6 year old Cuda version, then maybe yes. You might have to build a newer Python version and a whole bunch of Python libs for that, and this is quite time consuming. Cuda is not upgradable, and you are going to hit roadblocks.
Especially as a beginner you should use a supported platform. I can’t recommend Jetson Nano nowadays.
Hailo is a nice platform IF your software and models are compatible with this ASIC. NVidia GPUs are fully programmable and quite flexible. The Hailo ASICs are hardwired for certain types of operation. If this is what your software needs - good. If not - waste of money.
Ok thanks
I wasn’t aware of the Jetson Nano’s limitations I thought it was still a viable option despite being a bit outdated. You’re right it might not be the best choice
For the Hailo AI Kit, since YOLO is supported, it seems like a good fit for my needs, especially as I should be able to fine-tune YOLO with my own data
I don’t have much experience in this area, so I can’t say how easy it will be. Maybe it’s worth trying and seeing how it goes. If it doesn’t work out, I could consider the Jetson Orin Nano, but the price difference is significant (Raspberry Pi 5 + Hailo AI Kit = $150, Jetson Orin Nano Developer Kit = $500)
Go for the Raspberry Pi + Hailo AI kit if your software supports it.
We do have projects where we use a big Orin AGX 64GB with two of the bigger Hailo modules. Ok, this is a 2000$ system. The Orin AGX GPU does image postprocessing for two cameras, each of the Hailo modules does object recognition for one camera. So Jetson plus Hailo can also make sense, but not necessarily for a student.
Thanks,
For economic reasons, I’ll go with the Raspberry Pi + Hailo AI Kit it seems more practical for my budget. However, I’m curious about your setup with the Orin AGX and Hailo modules. Do you use the Jetson or the Hailo for coding the AI part, or is it a mix of both?
I’m asking because if I eventually decide to switch to a Jetson Orin Nano + Hailo setup, it seems like the code might be more easily adaptable if most of the work is already done on the Hailo
We can use both Jetson and Hailo for AI. Hailo only uses int8 data type while Jetson can use float and int datatypes and 8/16/32 bit datatypes. Using larger data types improves accuracy at the cost of speed and memory.
For that particular project we use Hailo for AI since the Orin GPU is busy with image decoding and processing. These are commercial projects so cost is not an issue.
ok thanks for your answer