We are excited to announce a new release for the Hailo Raspberry Pi AI Kit, packed with powerful new features that will enhance your AI development experience on Raspberry Pi 5. Whether you’re an embedded AI enthusiast or a seasoned developer, this release offers several exciting updates.
What’s New?
Hailo Python API Support
The Hailo Python API is now officially available for Raspberry Pi 5! This API allows you to easily run AI inference on the Hailo-8L AI processor using Python, simplifying the integration of deep learning models into your embedded applications. Explore Python code examples here.
CLIP Zero-Shot Classification App
We are introducing a zero-shot classification app based on the CLIP (Contrastive Language-Image Pretraining) model. This application allows real-time image recognition, matching video frames to relevant text prompts without needing pre-training for specific objects. It’s a powerful tool for AI tasks where flexibility is key. Check out the Hailo-CLIP GitHub repository for more information. Watch the demo on YouTube.
picamera2 Examples
The latest release also includes new picamera2 examples that showcase integration between Raspberry Pi’s camera stack and the Hailo AI processor. These examples demonstrate how to utilize the Hailo-8L for advanced AI-powered applications like object detection and pose estimation with the picamera2 library. Get started with picamera2 examples here.
Best Entry Point for All Updates
For a full list of updates and to get the latest features for your Raspberry Pi AI Kit, be sure to check out the Hailo Raspberry Pi 5 Examples Repository. This repository is your go-to source for all updates, new features, and examples for running AI on Raspberry Pi with the Hailo-8L AI processor.
In my opinion, the examples for Raspberry Pi5 should include standard support for IP cameras - RTSP streams, many people have asked about it on the forum and there are working script modifications there for example here: Using RTSP stream in Raspberry Pi5 - #6 by sanjoyg I think this would be very helpful for users without much experience.
Hey, nice to hear that support for the Hailo PiCamera2 is out, but I am getting an error on the first steps, when I am running the first example python3 pose.py
is there any hidden steps?
Traceback (most recent call last): File ‘/home/pi5/picamera2/examples/hailo/pose.py’, line 9, in from picamera2.devices import Hailo ImportError: cannot import name ‘Hailo’ from ‘picamera2.devices’ (/home/pi5/testenv/lib/python3.11/site-packages/picamera2/devices/init.py)"
Its a continuous troubleshooting instrument, i love it!
But what i think that is a weakness.
You have all software to let it run smoothly except for one thing to roll on a rpi5 and that is the converter.
The job should be this:
1st = Tensorflow
2nd = ONNX
3rd = HEF
The last part is tricky because you cannot place it on a rpi5 for it is not x86 based but arm64. All works except that. Any one have any pointers to put direct converter so i can decide cpu power an ai together. It is possible only the 3rd is a deadbeat.
I tried running the python scripts, unfortunately none of them worked for me. Of course i installed hailo-all.
I went into a venv for each demo script
python -m venv venv
source venv/bin/activate
and installed the resources and requirements.txt.
depth_estimation:
ModuleNotFoundError: No module named ‘hailo_platform’
detection_with_tracker:
python setup.py egg_info did not run successfully.
hailo_onnxruntime:
ERROR: Could not find a version that satisfies the requirement onnxruntime-openvino (from versions: none)
instance_segmentation:
ModuleNotFoundError: No module named ‘hailo_platform’
lane_detection:
ModuleNotFoundError: No module named ‘hailo_platform’
object_detection:
ModuleNotFoundError: No module named ‘hailo_platform’
pose_estimation:
ModuleNotFoundError: No module named ‘hailo_platform’
streaming:
ModuleNotFoundError: No module named ‘hailo_platform’
super_resolution:
espcnx4:
ModuleNotFoundError: No module named ‘hailo_platform’
srgan:
ModuleNotFoundError: No module named ‘hailo_platform’
Can you please tell me what i’m doing wrong?
Thanks
If you are using the Pi, it is installed on the the Pi’s “system” python as part of the hailo-all installation. However you will need to create the virtualenv with --system-site-packages flag to inherit it.
Are the Picamera2 examples supposed to work at this point in time?
I can’t get detect.py to work so far. There is a detection going on (after some fixes, for example the error about detection[4] being out of bounds), and an output, but the class_id is always empty/zero, which of course translates to a person every time. Looking at the hailo_output content, there’s really no value in there for the class of the detected object.
On the other hand, the rpicam-hello tool works perfectly. This is on a 26TOPS module.
Well, it works now for me. Putting this here if it helps anyone, but the only change that seems needed is this line in detect.py (picamera2 hailo examples):
for class_id, detections in enumerate(hailo_output):
Needs to be:
for class_id, detections in enumerate(hailo_output[0]):
Explore Python code examples here.