Hi all,
i’m wondering wihich picture details i should use for collecting data for a Fomo model to use with a ESP32-Cam board.
The details which i have in mind are
picture size, - is it better to get the pics at e.g. 240x240 size ore take it at a bigger size and scale it down ?
picture color, - take the pic at greyscale or color ?
The objects i want to dedect are are potato beetles (i’m working at a autonomus driving crawler which should collect it) and as a first stepp (because there are no potato beetles here atm …) dedect und count different little birds at the feeder.
I’d suggest you start by using a low resolution image and increase it slightly over time until you reach your desired accuracy. Please note that as you will increase the image size, the inference time, resource usage will also increase on your hardware.
For the ESP32-CAM with FOMO, I have seen that usually a 96x96 frame is a good trade-off but it always depends on the use case.