search

mit-han-lab / tinyengine

[NeurIPS 2020] MCUNet: Tiny Deep Learning on IoT Devices; [NeurIPS 2021] MCUNetV2: Memory-Efficient Patch-based Inference for Tiny Deep Learning; [NeurIPS 2022] MCUNetV3: On-Device Training Under 256KB Memory
https://mcunet.mit.edu
MIT License
809 stars 132 forks source link

mcunet-10fps inference failing shows nonperson always #18

Closed brainchip-india closed 1 year ago

brainchip-india commented 2 years ago

With the help of the tinyengine/tutorial I am able to run the mcunet_5ps(default) and able to see the inference is working fine, so tried changing model to mcunet_10fps in tinyengine/tutorial/examples/vww.py file able to run the code but the inference is failing (always showing no person though the person is present).

anything I need to take care to run mcunet_10fps? @meenchen @RaymondWang0 please help, thanks in advance.

mcunet_10fps_failed

brainchip-india commented 2 years ago

@meenchen @RaymondWang0 not sure you both have seen the issue, kindly reminding the issue.

meenchen commented 2 years ago

Hi @brainchip-india,

In the tutorial, the application code assumes the resolution of the model is 80x80, but for mcunet_10fps, it is actually 64x64. You will need to change the data loading part in the main.cpp for different models.

brainchip-india commented 2 years ago

Thanks @meenchen it worked, it's better you log it somewhere in the Readme for newbies like me, if its already logged please point me to the page because I would like to test different models thanks.