Closed absqh closed 2 months ago
We use the Tiny version of changemamba (see CDMamba/models/mamba_cd.py for details). To make a fair comparison, we unified the model into our framework for the experiment, without using any image enhancement methods, which may have led to different results. By the way, what can you achieve with F1 in the LEVIR-CD+ dataset?
The small model I use can reach 81.7%
If you are interested, try our framework, which integrates many change detection methods at the same time. If you find my answer helpful, please give us a star. Thanks!
Hello, I have a question. Why is the result so different from the original result of changemamba when comparing with changemamba? I used the changemamba model without pre-trained weights to run the LEVIR-CD+ dataset and there is also a difference between yours and the results