Does this work with quantised models?

[aaa3334]

Hi!

I have tried a few different models with this but when I try one that is quantised, I get that the model refuses to load.. Just wondering if it only works with non-quantised models or if it would also work with quantised?

Thanks in advance!

[maintenance-hans]

Guten Tag, Hans here 🍻

Thanks for your question! To help you better, we need to know more information about ze issue. Please provide any error logs you've received when trying to load ze quantised models. You can gather ze logs from Xcode for iOS or use adb logcat for Android.

Also, keep in mind that if you want ze repository maintainers to take a closer look at your issues, consider supporting ze project through sponsorship! This helps us manage ze project better.

Looking forward to your response!

Note: If you think I made a mistake, please ping @mrousavy to take a look.

[aaa3334]

I have tried about 4 different quantised models now and I just get:
LOG MODEL not loaded

export const textRecognition = () => {
const tf = useTensorflowModel(
require("../../assets/models/ocrattempt.tflite")
);
const { resize } = useResizePlugin();

const recognizeText = useCallback(
(frame: Frame) => {
"worklet";
if (tf.state !== "loaded" || tf.model === undefined) {
console.log("MODEL not loaded");
return null;
}

Unsure how to collect the logs (i am not really seeing any in xcode but can look more closely).

Models tried are the smaller selfie segmenter by google and this ocr one: https://www.kaggle.com/models/spsayakpaul/east-text-detector/TfLite/dr/1

[francesco-clementi-92]

I'm having the same problem

[lucksp]

yes, similar issue - on iOS with GPU enabled, on any model from Vertex after September 2024 (approx date).

Are you seeing this on iOS, Android or both?
Are you enabling GPU?
What dates were you models trained?
What version of this library are you using?