Training with high resolution images #1064
Description
Prerequisite
- I have searched the existing and past issues but cannot get the expected help.
- I have read the FAQ documentation but cannot get the expected help.
- The bug has not been fixed in the latest version.
🐞 Describe the bug
Input resolution is very important when a network is designed such that it will have the appropriate receptive field. Changing the input resolution (either decrease or increase) will lead to the change in receptive field also. If that is the case, then how come the network produces better accuracy by training with higher resolution images esp for smaller objects?
And also, when i train with higher resolution images, is it good idea to initialize the network with pre-trained weights that are trained with 640x640 or with some other resolution?
Environment
NA
Additional information
No response