Battle of the esp32 cams: XIAO esp32s3 Vs. AI Thinker

Seed Studio sent me a sample of their Seed Studio esp32 S3 Zao sense camera and that's quite a mouthful and I'm going to do run some tests to compare this with the AI thinker which I usually use now this retails I think for about 14 US Dollars which is about the same as the AI thinker on Amazon and then about half that on AliExpress so the first thing I notice about this is it comes configured for this external antenna which you have to press on and now to attach the external antenna to the AI thin car you have to desolder and resolder a surface mounted resistor now I can't tell you how many of these I've destroyed by trying to how many of these things I've destroyed by trying to desolder and raise soda on the surface mount resistor so I was pretty happy with that and the other thing I noticed about it is it comes with this USBC connector so you can Flash it more easily whereas for the for the AI think car you need to use like this ftdi kind of thing which is a bit of a which could be a bit of a hassle and so now I'm gonna just I've loaded on a sketch on these a person a tensorflow lite person detection sketch on both of these and I'm gonna see see if there's any speed improvement from the esp32 S3 uh over the AI thinker so I'll show you that the next scene hey everyone so I've updated the person detection sketch that came with a tensorflow lite esp32 Library so that it streams onto a local network and on the left I've got it running on the RC Studios our esp32s3 sense and on the right I've got it running on the AI thinker and here it's showing the milliseconds since the last inference here it's showing the score that it's a human or that it's a person and this is the frames per second multiplied by 10 so it's 2.3 frames per second and 2.1 frames per second and we see that the seat studio is about 5.6 or 5.5 seconds per inference whereas the AI think is about 8.2 seconds so that's quite an improvement that's about two and a half seconds Improvement around about 40 percent Improvement on average and the reason I've multiplied the frames per second by 10 is because I'm converting floats to integers and it's just easier to work with um when it's multiplied by 10 . so we're seeing there's a pretty high score for the person detection sketch and I'm a person and now I'm just going to point it the other direction so we can see it go to a low score and we'll see how long that takes 34 percent and this one's still working 18 percent so I think we've got a clear winner here which is the Seed Studio uh given that it's faster and it comes with a external aerial configured and it's easier to flash thank you for watching and I'll see you in the next video

Channel: Jonathan R

Views: 6,098

Keywords: esp32, esp32-cam, esp32 cam, tflite, tensorflow, tensorflowlite, esp32s3 cam, XIAO esp32 cam, XIAO, person detection, Esp32s3

Id: gYQDGoVRObA

Length: 3min 32sec (212 seconds)

Published: Sat Aug 19 2023