Page 1 of 1

Object recognition with mobilenet

Posted: Thu Jul 15, 2021 9:41 am
by ael-mess
Hello,
I have been reading the open sourced documentation on GitHub of the esp-dl library.
And I would like to implement a recognition algorithm based on mobilenet, but I am not sure on how to proceed.

What does exactly

Code: Select all

dl_matrix3d_t *dl_matrix3duf_mobilenetxx()
? is it the CNN of the model or just one depthwise separable operation? What are the steps taken in implementing the pretrained mobilenet model? And how to transfer the original weights to the ESP32?

If anyone could help I would appreciate it a lot!
Thanks,

Re: Object recognition with mobilenet

Posted: Thu Jul 15, 2021 11:34 am
by yehangyang
Hi, aei-mess

ESP-DL is under refactoring. Please follow the new ESP-DL on branch esp-dl/v2_alpha temporarily. About how to customize your own model check https://github.com/espressif/esp-dl/tre ... a/tutorial

Best Regards.

Re: Object recognition with mobilenet

Posted: Thu Jul 15, 2021 2:13 pm
by ael-mess
Hello yehangyang,

Thank you for your fast response, good to here that the v2 of the est-dl is coming.
So you don't advise using the first release for new projects? I checked out the documentation of the v2_alpha and not all the layers are available. In my case, I will be missing the fully connected and softmax layers for the mobilenet.
Should I implement them by myself?

Regards,