search

espressif / esp-dl

Espressif deep-learning library for AIoT applications
MIT License
549 stars 118 forks source link

Latency of Face Recognition (AIV-689) #160

Closed assman closed 6 months ago

assman commented 6 months ago

The docs say that the latency for a 16bit S3 is more than that of an 8bit S3? Is it for real or is there a mistake here?

Screenshot 2024-03-29 at 3 39 32 PM
MichelBerg commented 6 months ago

Why should it be a mistake? As it says, the int8 model has a smaller footprint and shorter latency. Think about having approximate half the data compared to the int16 model, it has to be faster.

assman commented 6 months ago

That makes sense, I’m so sorry I mistook the 8bit and 16bit for something else.