I thought the available tools were too complicated. They didn't support the latest AI models, required bloated shell scripts to install, and lacked features and speed.
So, I built the best one. People can import their own models (if the architecture is supported), and it already supports the best models in the field, with more coming.
It's fairly efficient, the binary size of the .exe is just 30 MB, of which 10 MB is just the font to support multiple languages. We also bundle the dependencies (ONNX-CUDA and ffmpeg) in case your computer doesn't have them. There is some SIMD in the pre-processing, and we'll add SIMD in the post-processing too.
Free and open source, just to help the world protect nature, from monitoring biodiversity to even early detection of wildfires.
It's being used in China, Japan, Europe, LATAM, USA. Everywhere in the world, really.
The repo: https://github.com/boquila/boquilahub/
Also, the same binary contains a minimal CLI.
The goal is that the same binary can work for:
- Big clusters that want to process thousands of cameras in real-time.
- Tiny Raspberry Pis that want to do inference on-edge.
Isn't that cool?