Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

How to improve the results #47

Open
NasrinR791 opened this issue Dec 11, 2021 · 0 comments
Open

How to improve the results #47

NasrinR791 opened this issue Dec 11, 2021 · 0 comments

Comments

@NasrinR791
Copy link

Hi @andreas128
Thanks for your great work and sharing the code.
I tried to train SRFlow using REDS dataset. After training the RRDBN network on REDS (with PSNR around 29 dB), I used this config to train SRFlow. However, the resulted SR images are not good and the PSNR is around 17 dB.
I did not change the code by the way. Could you please give me a hint on how can I improve the training?

network_G:
which_model_G: SRFlowNet
in_nc: 3
out_nc: 3
nf: 64
nb: 25
upscale: 4
train_RRDB: false
train_RRDB_delay: 0.5

flow:
K: 10
L: 3
noInitialInj: true
coupling: CondAffineSeparatedAndCond
additionalFlowNoAffine: 2
split:
enable: true
fea_up0: true
stackRRDB:
blocks: [ 1,5, 15, 22 ]
concat: true

train:
manual_seed: 1386
lr_G: !!float 1e-5
weight_decay_G: 0
beta1: 0.9
beta2: 0.99
lr_scheme: MultiStepLR
warmup_iter: -1 # no warm up
lr_steps_rel: [ 0.5, 0.75, 0.9, 0.95 ]
lr_gamma: 0.8
weight_l1: 0.1

niter: 300000
val_freq: !!float 5e3

validation settings

val:
heats: [ 0.0, 0.5]
n_sample: 1

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant