how can I reach or close to the performance in the papper by training with single GPU?

[boqihan]

I can't get the performance of SBD dataset in the papper with single GPU, with the code 'python train.py --dataset sbd --model dff --backbone resnet101 --checkname dff --base-size 352 --crop-size 352 --epochs 10 --batch-size 4 --lr 0.05 --workers 8'. There is a huge gap to the performance in the papper.
Is there something needed to be modify? Due to hardware reasons, I can only set batchsize to 4. Should I modify other parameters?

[boqihan]

Of course. I had met this problem, but unfortunately I frogot it. Sorry for that. To be honest, after several times experients, I thought the provided codes can not be reimplent. So maybe try something new!

…

------------------ 原始邮件 ------------------ 发件人: "Lavender105/DFF" ***@***.***>; 发送时间: 2021年5月28日(星期五) 晚上8:00 ***@***.***>; ***@***.******@***.***>; 主题: Re: [Lavender105/DFF] how can I reach or close to the performance in the papper by training with single GPU? (#23) hello, May I ask you a question ? I met a problem, in parallel.py, line 160, “output = module(*(input + target), **kwargs)” TypeError: can only concatenate list (not "tuple") to list， could you please tell how to slove it — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[cannonli7]

Of course. I had met this problem, but unfortunately I frogot it. Sorry for that. To be honest, after several times experients, I thought the provided codes can not be reimplent. So maybe try something new!
…
------------------ 原始邮件 ------------------ 发件人: "Lavender105/DFF" @.>; 发送时间: 2021年5月28日(星期五) 晚上8:00 @.>; @.@.>; 主题: Re: [Lavender105/DFF] how can I reach or close to the performance in the papper by training with single GPU? (#23) hello, May I ask you a question ? I met a problem, in parallel.py, line 160, “output = module(*(input + target), **kwargs)” TypeError: can only concatenate list (not "tuple") to list， could you please tell how to slove it — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

Thank you