Closed HaoZhang1018 closed 11 months ago
Hi, thanks for your interest. Sorry we have not verified our methods by using RTX 3090 GPUs and the batch size to 3. Maybe a larger epoch or different training strategy could be helpful. We appreciate that you provide this result reproduced from your side. Thanks.
Thank you for this excellent work. I retrain CMNext on the MFNet dataset, but I am unable to achieve the scores reported in the original paper (I obtained mIoU: 57.63). The only difference is that I am using 4 RTX 3090 GPUs and modified the batch size to 3 due to memory constraints. Could you provide some advice to help me reproduce the reported scores (59.9)?