Training Process Issues

[xiaoshuai0719]

Hello author, Thank you for your work.I encountered some issues during training. In the yml dataset section, I only kept the training dataset, and I commented out the sections for the validation datasets. May I ask whether in the yaml section I should keep one validation set or maintain the code settings for all validation set parts during the training process?Here are the parts that I commented out during training:

val_snow_s:

name: ValSet_Snow100K-S

type: Dataset_PairedImage

dataroot_gt: "E:/Snow100K-Test/Snow100K-S/gt"

dataroot_lq: "E:/Snow100K-Test/Snow100K-S/synthetic"

io_backend:

type: disk

val_snow_l:

name: ValSet_Snow100K-L

type: Dataset_PairedImage

dataroot_gt: /home1/ssq/data/allweather/test/Snow100K-L/gt/

dataroot_lq: /home1/ssq/data/allweather/test/Snow100K-L/synthetic/

io_backend:

type: disk

val_test1:

name: ValSet_Test1

type: Dataset_PairedImage

dataroot_gt: /home1/ssq/data/allweather/test/Test1/gt/

dataroot_lq: /home1/ssq/data/allweather/test/Test1/input/

io_backend:

type: disk

val_raindrop:

name: ValSet_RainDrop

type: Dataset_PairedImage

dataroot_gt: /home1/ssq/data/allweather/test/RainDrop/gt/

dataroot_lq: /home1/ssq/data/allweather/test/RainDrop/input/

io_backend:

type: disk

[sunshangquan]

Hi @xiaoshuai0719 , thank you for your attention. The validation sets can be changed by also changing Lines 115-162 in the train.py. If you want to remove the entire validation process, you could set whatever your validation set is (for example a single identical image for both input and ground-truth) and change the value of 60 here to 0 to skip validation process. If this cannot address your issue, could you further show the specific error information you encountered?

[xiaoshuai0719]

Hello author, I would like to know if I need to comment out the validation dataset part during the process of running train.py, or should I only leave the single validation set that I need, in other words, do I need to use the validation set val during the training process?

[xiaoshuai0719]

In other words, do I need to comment out the val dataset settings in the Allweather_Histoformer.yml during the training process?

[xiaoshuai0719]

When I first started training, I commented out the 'val' section in the Allweather_Histoformer.yml file (as shown in the figure below), so for normal training, I should also ensure that the val dataset is loaded properly without commenting out, so that I can get the best.pth, right?

[sunshangquan]

Indeed, you may safely try it and comment it out for your intended purpose.