Are you sure you want to delete this access key?
This tutorial explains how to control your image size when training on your own data.
There are 3 hyperparamters control the training size:
There is 1 hyperparameter constrols the testing size:
The self.input_size is suggested to set to the same value as self.test_size. By default, it is set to (640, 640) for most models and (416, 416) for yolox-tiny and yolox-nano.
When training on your custom dataset, you can use multiscale training in 2 ways:
【Default】Only specifying the self.input_size and leaving others unchanged.
If so, the actual multiscale sizes range from:
[self.input_size[0] - self.multiscale_range*32, self.input_size[0] + self.multiscale_range*32]
For example, if you only set:
self.input_size = (640, 640)
the actual multiscale range is [640 - 5*32, 640 + 5*32], i.e., [480, 800].
You can modify self.multiscale_range to change the multiscale range.
Simultaneously specifying the self.input_size and self.random_size
self.input_size = (416, 416)
self.random_size = (10, 20)
In this case, the actual multiscale range is [self.random_size[0]*32, self.random_size[1]*32], i.e., [320, 640]
Note: You must specify the self.input_size because it is used for initializing resize aug in dataset.
If you want to train in a single scale. You need to specify the self.input_size and self.multiscale_range=0:
self.input_size = (416, 416)
self.multiscale_range = 0
DO NOT set the self.random_size.
Press p or to see the previous file or, n or to see the next file
Are you sure you want to delete this access key?
Are you sure you want to delete this access key?
Are you sure you want to delete this access key?
Are you sure you want to delete this access key?