mrq/DL-Art-School
1
3
Fork 5
Code Issues Pull Requests Packages Projects Releases Wiki Activity
fc3ec8e3a2
DL-Art-School/codes/options/train/train_ESRGAN.yml

83 lines
1.6 KiB
YAML
Raw Normal View History

mmsr
2019-08-23 13:42:47 +00:00
#### general settings
name: 003_RRDB_ESRGANx4_DIV2K
use_tb_logger: true
model: srgan
distortion: sr
scale: 4
Change GT_size to target_size
2020-04-22 06:37:41 +00:00
gpu_ids: [0]
Fixup upconv for the next attempt!
2020-05-02 01:56:14 +00:00
amp_opt_level: O1
mmsr
2019-08-23 13:42:47 +00:00
#### datasets
datasets:
train:
name: DIV2K
mode: LQGT
Fixup upconv for the next attempt!
2020-05-02 01:56:14 +00:00
dataroot_GT: E:/4k6k/datasets/div2k/DIV2K800_sub
dataroot_LQ: E:/4k6k/datasets/div2k/DIV2K800_sub_bicLRx4
mmsr
2019-08-23 13:42:47 +00:00
use_shuffle: true
Change GT_size to target_size
2020-04-22 06:37:41 +00:00
n_workers: 16 # per GPU
Implement ResGen arch This is a simpler resnet-based generator which performs mutations on an input interspersed with interpolate-upsampling. It is a two part generator: 1) A component that "fixes" LQ images with a long string of resnet blocks. This component is intended to remove compression artifacts and other noise from a LQ image. 2) A component that can double the image size. The idea is that this component be trained so that it can work at most reasonable resolutions, such that it can be repeatedly applied to itself to perform multiple upsamples. The motivation here is to simplify what is being done inside of RRDB. I don't believe the complexity inside of that network is justified.
2020-05-05 17:59:46 +00:00
batch_size: 16
Change GT_size to target_size
2020-04-22 06:37:41 +00:00
target_size: 128
mmsr
2019-08-23 13:42:47 +00:00
use_flip: true
use_rot: true
color: RGB
val:
Change GT_size to target_size
2020-04-22 06:37:41 +00:00
name: div2kval
mmsr
2019-08-23 13:42:47 +00:00
mode: LQGT
Fixup upconv for the next attempt!
2020-05-02 01:56:14 +00:00
dataroot_GT: E:/4k6k/datasets/div2k/div2k_valid_hr
dataroot_LQ: E:/4k6k/datasets/div2k/div2k_valid_lr_bicubic
mmsr
2019-08-23 13:42:47 +00:00
#### network structures
network_G:
Implement ResGen arch This is a simpler resnet-based generator which performs mutations on an input interspersed with interpolate-upsampling. It is a two part generator: 1) A component that "fixes" LQ images with a long string of resnet blocks. This component is intended to remove compression artifacts and other noise from a LQ image. 2) A component that can double the image size. The idea is that this component be trained so that it can work at most reasonable resolutions, such that it can be repeatedly applied to itself to perform multiple upsamples. The motivation here is to simplify what is being done inside of RRDB. I don't believe the complexity inside of that network is justified.
2020-05-05 17:59:46 +00:00
which_model_G: ResGen
nf: 256
mmsr
2019-08-23 13:42:47 +00:00
network_D:
Implement ResGen arch This is a simpler resnet-based generator which performs mutations on an input interspersed with interpolate-upsampling. It is a two part generator: 1) A component that "fixes" LQ images with a long string of resnet blocks. This component is intended to remove compression artifacts and other noise from a LQ image. 2) A component that can double the image size. The idea is that this component be trained so that it can work at most reasonable resolutions, such that it can be repeatedly applied to itself to perform multiple upsamples. The motivation here is to simplify what is being done inside of RRDB. I don't believe the complexity inside of that network is justified.
2020-05-05 17:59:46 +00:00
which_model_D: discriminator_resnet_passthrough
nf: 42
mmsr
2019-08-23 13:42:47 +00:00
#### path
path:
Fixup upconv for the next attempt!
2020-05-02 01:56:14 +00:00
pretrain_model_G: ~
mmsr
2019-08-23 13:42:47 +00:00
strict_load: true
resume_state: ~
#### training settings: learning rate scheme, loss
train:
lr_G: !!float 1e-4
weight_decay_G: 0
beta1_G: 0.9
beta2_G: 0.99
lr_D: !!float 1e-4
weight_decay_D: 0
beta1_D: 0.9
beta2_D: 0.99
lr_scheme: MultiStepLR
niter: 400000
warmup_iter: -1 # no warm up
lr_steps: [50000, 100000, 200000, 300000]
lr_gamma: 0.5
Implement ResGen arch This is a simpler resnet-based generator which performs mutations on an input interspersed with interpolate-upsampling. It is a two part generator: 1) A component that "fixes" LQ images with a long string of resnet blocks. This component is intended to remove compression artifacts and other noise from a LQ image. 2) A component that can double the image size. The idea is that this component be trained so that it can work at most reasonable resolutions, such that it can be repeatedly applied to itself to perform multiple upsamples. The motivation here is to simplify what is being done inside of RRDB. I don't believe the complexity inside of that network is justified.
2020-05-05 17:59:46 +00:00
mega_batch_factor: 1
mmsr
2019-08-23 13:42:47 +00:00
pixel_criterion: l1
pixel_weight: !!float 1e-2
feature_criterion: l1
feature_weight: 1
Fixup upconv for the next attempt!
2020-05-02 01:56:14 +00:00
feature_weight_decay: .98
feature_weight_decay_steps: 500
feature_weight_minimum: .1
gan_type: gan # gan | ragan
mmsr
2019-08-23 13:42:47 +00:00
gan_weight: !!float 5e-3
Fixup upconv for the next attempt!
2020-05-02 01:56:14 +00:00
D_update_ratio: 2
mmsr
2019-08-23 13:42:47 +00:00
D_init_iters: 0
manual_seed: 10
Change GT_size to target_size
2020-04-22 06:37:41 +00:00
val_freq: !!float 5e2
mmsr
2019-08-23 13:42:47 +00:00
#### logger
logger:
Change GT_size to target_size
2020-04-22 06:37:41 +00:00
print_freq: 50
save_checkpoint_freq: !!float 5e2
Reference in New Issue Copy Permalink