Open hellodrx opened 3 years ago
same as you ,bro, my result and parameters is listed as you can see, the model is MARNet --epochs 100 - -dataset sha --loss 3avg-ms-ssim - -lazy_val 0 [2021-05-09 19:06:58,917], ## MARNet Epoch 100/100 Loss:1.693027, lr:0.0000025, [CUR]:71.1, 110.0, [Best]:68.7, 108.1
i wonder whether this is the problem of some parameters that are not set probably
hi, I got the listed results with default parameters in train_generic.py. Maybe you can adjust the lr, or check if there is some error in the testing process.
the lr is set as default 1e-5 with step optimizer, this res is my test on sha, i would try it on qnrf 3qu very much
the lr is set as default 1e-5 with step optimizer, this res is my test on sha, i would try it on qnrf 3qu very much
Did you get any progress?
the lr is set as default 1e-5 with step optimizer, this res is my test on sha, i would try it on qnrf 3qu very much
Did you get any progress?
the lr is set as default 1e-5 with step optimizer, this res is my test on sha, i would try it on qnrf 3qu very much
Did you get any progress?
Not yet.
what are these values in CUR and BEST MAE and RMSE ? and what CUR stands for
Hi, Liangzi,
Firstly, thanks for your wonderful work. I ran your code as the guideline suggested, however, I cannot get proper results.
On QNRF, I get the following result, [2021-05-02 23:55:37,254], ## MARNet Epoch 100/100 Loss:1.257166, lr:0.0000025, [CUR]:111.3, 185.2, [Best]:95.2, 167.8 which is far away from what the paper and this repo have claimed.
On SHA, the result is [2021-05-04 08:16:29,637], ## MARNet Epoch 300/300 Loss:1.239268, lr:0.0000002, [CUR]:68.0, 110.4, [Best]:65.9, 106.9
Could you kindly help me out