About result quality

[lidawenwenwen]

Great work! I tried to reproduce the results using the data mentioned in the README, but the result quality doesn’t seem very good. I’m using the Tanks dataset and running the bash scripts/train_tnt.sh script. I also tried the other datasets mentioned in the README, but the results were similarly not very good.

Is there anything else I need to modify?

[lidawenwenwen]

@linjohnss

[chuhantong]

I also tested this scene, but the metrics don't seem to be very good.