How to apply this model to large-sized images?

[crystallo3921]

Thank you so much for your remarkable contributions to the field of change detection! I have successfully trained and inferred this model using the provided samples. However, I have a question regarding the inference of changes between two large-sized images, such as high-resolution satellite or aerial images. Could you please advise me on how to apply this model in such cases?

I have previously employed the sliding window strategy, which enabled me to successfully infer changes in a pair of high-resolution aerial images. However, in this particular scenario, all the configurations are written in files, and the training and testing processes are automated using Runner. I have carefully reviewed the mmengine documentation, but I am still a bit confused, especially regarding the adjustment of configurations. Could you kindly provide some guidance and insights? Your assistance is greatly appreciated. Thank you so much!

[likyoo]

Thanks for your attention, you can refer here.

[crystallo3921]

Thanks for your attention, you can refer here.

Thank you so much for your prompt reply. I would like to provide more specific details regarding my question. When I modify the line from 'test_cfg=dict(mode='whole')' to 'test_cfg=dict(mode='slide', crop_size=crop_size, stride=(crop_size[0]//2, crop_size[1]//2))', I would like to know how the remaining test configuration parameters should be adjusted accordingly.

Because I intend to switch the test dataset to two large-sized images,without labels, what changes should I make to the test_dataset parameters? The origin test_dataset in config is:

test_dataloader = dict( batch_size=1, num_workers=4, persistent_workers=True, sampler=dict(type='DefaultSampler', shuffle=False), dataset=dict( type='LEVIR_CD_Dataset', data_root='data/LEVIR-CD', data_prefix=dict( seg_map_path='test/label', img_path_from='test/A', img_path_to='test/B'), pipeline=[ dict(type='MultiImgLoadImageFromFile'), dict(type='MultiImgResize', scale=(1024, 1024), keep_ratio=True), dict(type='MultiImgLoadAnnotations'), dict(type='MultiImgPackSegInputs') ]))

Should I modify the dataset type（'LEVIR_CD_Dataset'） or create a customized dataset?

Additionally, could you please advise on any other parameters that are relevant to the test section and may require modification?