kuasar VS keras-ocr

Right now most early examples alas boot a container with a wasm runtime for each wasm instance, which is a sad waste. The whole advantage of wasm should be very lightweight low overhead wasm runtime instances atop a common wasm process. Having a process or container for each instance loses a ton of the benefit, makes it not much better than a regular container.

Thankfully there is work like the Containerd Sandbox API which enables new architectures like this. https://github.com/containerd/containerd/issues/4131

It's still being used to spawn a wasm processes per instance for now, but container runtime project Kuasar is already using the Sandbox API to save significant resources, and has already chimed in in comments on HN to express a desire to have shared-process/multi-wasm-instamxe runtimes, which could indeed allow sub ms spawning that could enable instance per request architectures. https://github.com/kuasar-io/kuasar

I'm trying to detect things like speed limits, stop signs, retail building signs from a relatively low quality dashcam. The video is 1440p, but the optics aren't great.

So far I've been using generic OCR models like [1] and [2], but the results aren't great.

[1] https://github.com/faustomorales/keras-ocr

Yes, you should be able to do it by replacing the the backbone and training the other parts again. The results may be better or worse than you expected. See: https://github.com/faustomorales/keras-ocr/issues/113

There's plenty of OCR models around. If you'll have a Python server handling this, you can try keras-ocr. If you want to do this right in the browser, you can use a tflite model.