Hi, thanks for your great work and for releasing the code.
Could you share the pretrained model weights?
I haven’t been able to reproduce the results reported in the paper using the provided code.
The best mIoU I can reach is around 76.49%, which is significantly lower than the reported 85.08% on Light4Mars-B.
Is there any specific training detail or reason that might explain why my reproduction falls short of the expected result?
The model was trained on single rtx 4090 24gb.
Thanks.
Hi, thanks for your great work and for releasing the code.
Could you share the pretrained model weights?
I haven’t been able to reproduce the results reported in the paper using the provided code.
The best mIoU I can reach is around 76.49%, which is significantly lower than the reported 85.08% on Light4Mars-B.
Is there any specific training detail or reason that might explain why my reproduction falls short of the expected result?
The model was trained on single rtx 4090 24gb.
Thanks.