Input resolution

[EthanAbitbol3]

问题确认 Search before asking

• 我已经搜索过问题，但是没有找到解答。I have searched the question and found no related answer.

请提出你的问题 Please ask your question

Hello and thank you for your work!
I had a question about the input resolution of an algorithm. I am currently using ade20k which has an image size that varies but for example on segformer it is mentioned that the input size is 512x512, how is this done?

thanks again !

[Asthestarsfalll]

By using image interpolation algorithms, which are available in most Python image libraries under the name "resize." Some deep learning frameworks may refer to these algorithms as "interpolate."

[EthanAbitbol3]

Thank you for your help and your time !

for example in this configuration file :

base: '../base/cityscapes_1024x1024.yml'

batch_size: 2
iters: 160000

model:
type: SegFormer
backbone:
type: MixVisionTransformer_B0
pretrained: https://bj.bcebos.com/paddleseg/dygraph/backbone/mix_vision_transformer_b0.tar.gz
embedding_dim: 256
num_classes: 19

optimizer:
inherited: False
type: AdamW
beta1: 0.9
beta2: 0.999
weight_decay: 0.01

lr_scheduler:
type: PolynomialDecay
learning_rate: 0.00006
power: 1

loss:
types:
- type: CrossEntropyLoss
coef: [1]

test_config:
is_slide: True
crop_size: [1024, 1024]
stride: [768, 768]

Does the crop size specified in the test_config refer to the input resolution?

[Asthestarsfalll]

Does the crop size specified in the test_config refer to the input resolution?

Yes, the input size is [1024, 1024].
By the way, crop is another way to adjust the size of images. It does not alter the scale of image, but crop some part of images