125 points by gestrtech 5 months ago flag hide 20 comments
gnomeprogrammer 5 months ago next
This looks really promising! Has anyone tried implementing it in a real-world project yet?
yagni 5 months ago next
I'm excited to see how this can help reduce server-side computation and improve user experience.
neuralwizard 5 months ago prev next
I've been playing around with the API, and the accuracy of gesture recognition is impressive.
gnomeprogrammer 5 months ago next
NeuralWizard, could you share a simple example of how you're using this API for gesture recognition?
capybaradev 5 months ago prev next
How is the performance on lower-end devices? Is it memory-efficient?
techiecrowd 5 months ago prev next
Which ML model(s) are being used under the hood for gesture recognition?
yagni 5 months ago next
I believe it's using a variation of Convolutional Neural Networks (CNNs) optimized for this specific task.
mitdeveloper 5 months ago prev next
Are there any plans to introduce customization of the API and the models? This could be useful for special use cases.
neuralwizard 5 months ago next
Great question! They've mentioned that customization options will be rolled out in the near future.
turingtested 5 months ago prev next
Will it be possible to contribute or build custom models based on the API? Would be interesting to see a community-driven project for special use cases.
marshall_stack 5 months ago next
The team behind this project has indicated in their documentation that they are open to community contributions.
inspiredbot 5 months ago prev next
How noisy does the data have to be for this API to fail in recognizing gestures? Can you provide some analysis on this matter?
deepneuralthinker 5 months ago next
The API's robustness depends on the quality of the input it receives. In most cases, it handles mild noise pretty well, but performance may degrade under high noise conditions.
makingwaves 5 months ago prev next
How much preprocessing of on-device data is required before this gesture recognition API can effectively work?
neuralwizard 5 months ago next
The data quality is crucial, but usually, a simple normalization step is sufficient before sending your data to the API.
letrainsoftware 5 months ago prev next
I noticed this API requires an internet connection. Are there plans for an offline version in the future?
hmartin17 5 months ago next
If you require offline gesture recognition, there are well-documented alternatives with cuirosity, OpenCV and TensorFlow-lit. However, their performance might vary.
algorithmsareus 5 months ago prev next
The doesn't seem very viable for privacy-preserving applications. Has anyone found a solution to address this concern?
haskellshaman 5 months ago next
You can run this API locally, but it still requires internet connectivity for the initial model download. Running a local instance with local modifications might help in privacy-preserving applications.
mindmeld 5 months ago prev next
I just started trying out the demo on their website. Even on a not-so-recent device, it recognizes gestures remarkably well. Kudos to the developers!