Hello everyone,
Does anyone know of a specific model compatible with Hailo to detect chess pieces and possibly the chessboard with it? Are there any tutorials for adapting it with Hailo and a camera, as it seems to me normally one of the yolo models uses the chess piece detection class.
I would recommend to check out the following two pages:
Our Model Zoo retraining will allow you to easily retrain a Yolo model.
GitHub - Model Zoo - Retrain on custom dataset
Kaggle or some other website can provide you with a dataset you can use for training. It looks like there are some dataset related to chess.
Kaggle
Hello @KlausK
Thank you for your reply.
I was indeed able to convert the best.pt trained chess model to ONNX format and I understood that to convert it to .hef, I had to install the Hailo AI - Docker Software Suite but on Ubuntu (to then be able to use the hailomoz command to convert), but isn’t there a way to do that on Debian 12 which runs on my rpi5? If not, are there any other alternatives for quickly converting .onnx to .hef on rpi5?
The Hailo-8 can perform model inference much more efficiently than a typical GPU. One key reason is that it uses integer computation rather than floating-point, which significantly reduces power consumption. To enable this, models must first be quantized - a process that converts floating-point parameters to integers. Achieving this with minimal loss in accuracy is challenging and often requires advanced algorithms and, in some cases, a powerful GPU.
While the Raspberry Pi is not suitable for performing model training or quantization, it excels as a platform for running inference when paired with the Hailo-8 accelerator.