Poor performance using custom dataset results

[yuanjie0217]

I am trying to use a custom data set to train on the instant-ngp project. My data set is simulated building data. This is simulated based on the SAR imaging principle. I added a mask to the simulation results to make fewer impurities. Then I estimated the camera pose through colmap and made my own data set with reference to the project's json file. However, the performance of my data set on instant-ngp was very poor. How can I improve it to improve the final result?

[yuanjie0217]

This is the final rendering result