The above code is the implementation of OIPF-SCT. The corresponding dataset and weights can be found at the following links: https://pan.baidu.com/s/1GZuwZfKDu0bPV4ewCVGOxQ?pwd=52u4 password:52u4

Importantly, we would like to reaffirm the effectiveness of our approach. Taking the experimental results on the NUDT-SIRST dataset as an example, the IoU metric stated in the paper "SCTransNet" published in TGRS in April 2024, is 94.09%. Following the same configuration as the authors, we reproduced the metric and obtained a result of 93.08%. However, after improving the model with OIPF, the metric increased to 95.43%.

Furthermore, to the best of our knowledge, 95.43% is a highly competitive result on the NUDT-SIRST dataset. For instance, the method "Pick of the Bunch" published in TGRS in September 2024, reported an IoU of 94.28%.

The code related to OIPF is available at https://github.com/Kingwin97/OIPF-Net, and the weights corresponding to the 95.43% result are also publicly accessible. You can obtain the same metrics by running our test file test_twoflow.py, or you can retrain the model using train_twoflow.py.