Segmentation results

[sheshap]

I tried to recreate segmentation results but could achieve only
Best mIoU(c): 0.811, Best mIoU (i): 0.846

Please let me know if any changes need to be done to achieve the results in the paper
i.e. mIoU(c): 0.821, mIoU (i): 0.851

Thanks in Advance.

[chih-hao-lin]

I think you can try to run the experiments with more epochs (about 150), or load the trained model weights and train again with a smaller learning rate (= 5e-5).
Hope this helps.