I have a question about the data set for classification:
I put an esp32 cam in front of our door and I want to know if the door is open or close.
the camera is pretty far from the doors (5-6 meters) so it also pictures some background.
The camera will remain in the same place always.
What is the best way to collect data set?
Should I take a picture in a high resolution just for the data set? should I crop each image and remove the background?
Should I take pictures at the same resolution that I’ll take after I’ll teach the model (low resolution)?
Can someone advise me on how to get the best model?
It is usually advisable to use the target deployment device to collect images for your dataset for even better performance during inference, if you are in a position to. So it would be great if you could collect your dataset using the esp32 cam. You can however add the high resolution ones if you find your dataset is too small.
Also since during deployment your camera will still capture the background, you might not necessarily want to crop the images. This way your model will learn how to differentiate the background by itself while still able to tell whether the door is closed or open.