eschmidbauer/DL-Art-School
1
0
Fork 0
forked from mrq/DL-Art-School
Code Issues Pull Requests Packages Projects Releases Wiki Activity
602f86bfa4
DL-Art-School/codes/options/train/train_ESRGAN_blacked_xl.yml

84 lines
1.7 KiB
YAML
Raw Normal View History

Add support for passthrough disc/gen Add RRDBNetXL, which performs processing at multiple image sizes. Add DiscResnet_passthrough, which allows passthrough of image at different sizes for discrimination. Adjust the rest of the repo to allow generators that return more than just a single image.
2020-05-04 20:01:43 +00:00
#### general settings
name: blacked_fix_and_upconv_xl
use_tb_logger: true
model: srgan
distortion: sr
scale: 4
gpu_ids: [0]
amp_opt_level: O1
#### datasets
datasets:
train:
name: vixcloseup
mode: LQGT
Random config changes
2020-05-06 23:25:48 +00:00
dataroot_GT: E:\4k6k\datasets\4k_closeup\hr
dataroot_LQ: [E:\4k6k\datasets\4k_closeup\lr_corrupted, E:\4k6k\datasets\4k_closeup\lr_c_blurred]
Add support for passthrough disc/gen Add RRDBNetXL, which performs processing at multiple image sizes. Add DiscResnet_passthrough, which allows passthrough of image at different sizes for discrimination. Adjust the rest of the repo to allow generators that return more than just a single image.
2020-05-04 20:01:43 +00:00
doCrop: false
use_shuffle: true
Random config changes
2020-05-06 23:25:48 +00:00
n_workers: 12 # per GPU
batch_size: 15
Add support for passthrough disc/gen Add RRDBNetXL, which performs processing at multiple image sizes. Add DiscResnet_passthrough, which allows passthrough of image at different sizes for discrimination. Adjust the rest of the repo to allow generators that return more than just a single image.
2020-05-04 20:01:43 +00:00
target_size: 256
color: RGB
val:
name: adrianna_val
mode: LQGT
dataroot_GT: E:\4k6k\datasets\adrianna\val\hhq
dataroot_LQ: E:\4k6k\datasets\adrianna\val\hr
#### 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
Random config changes
2020-05-06 23:25:48 +00:00
nb_denoiser: 20
nb_upsampler: 10
Add support for passthrough disc/gen Add RRDBNetXL, which performs processing at multiple image sizes. Add DiscResnet_passthrough, which allows passthrough of image at different sizes for discrimination. Adjust the rest of the repo to allow generators that return more than just a single image.
2020-05-04 20:01:43 +00:00
network_D:
which_model_D: discriminator_resnet_passthrough
Random config changes
2020-05-06 23:25:48 +00:00
nf: 64
Add support for passthrough disc/gen Add RRDBNetXL, which performs processing at multiple image sizes. Add DiscResnet_passthrough, which allows passthrough of image at different sizes for discrimination. Adjust the rest of the repo to allow generators that return more than just a single image.
2020-05-04 20:01:43 +00:00
#### path
path:
Random config changes
2020-05-06 23:25:48 +00:00
#pretrain_model_G: ../experiments/pretrained_resnet_G.pth
Fix megabatch scaling, log low and med-res gen images
2020-05-05 14:34:57 +00:00
#pretrain_model_D: ~
Add support for passthrough disc/gen Add RRDBNetXL, which performs processing at multiple image sizes. Add DiscResnet_passthrough, which allows passthrough of image at different sizes for discrimination. Adjust the rest of the repo to allow generators that return more than just a single image.
2020-05-04 20:01:43 +00:00
strict_load: true
Random config changes
2020-05-06 23:25:48 +00:00
resume_state: ../experiments/blacked_fix_and_upconv_xl/training_state/2500.state
Add support for passthrough disc/gen Add RRDBNetXL, which performs processing at multiple image sizes. Add DiscResnet_passthrough, which allows passthrough of image at different sizes for discrimination. Adjust the rest of the repo to allow generators that return more than just a single image.
2020-05-04 20:01:43 +00:00
#### training settings: learning rate scheme, loss
train:
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
lr_G: !!float 1e-4
Add support for passthrough disc/gen Add RRDBNetXL, which performs processing at multiple image sizes. Add DiscResnet_passthrough, which allows passthrough of image at different sizes for discrimination. Adjust the rest of the repo to allow generators that return more than just a single image.
2020-05-04 20:01:43 +00:00
weight_decay_G: 0
beta1_G: 0.9
beta2_G: 0.99
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
lr_D: !!float 2e-4
Add support for passthrough disc/gen Add RRDBNetXL, which performs processing at multiple image sizes. Add DiscResnet_passthrough, which allows passthrough of image at different sizes for discrimination. Adjust the rest of the repo to allow generators that return more than just a single image.
2020-05-04 20:01:43 +00:00
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: [20000, 40000, 50000, 60000]
lr_gamma: 0.5
Random config changes
2020-05-06 23:25:48 +00:00
mega_batch_factor: 3
Add support for passthrough disc/gen Add RRDBNetXL, which performs processing at multiple image sizes. Add DiscResnet_passthrough, which allows passthrough of image at different sizes for discrimination. Adjust the rest of the repo to allow generators that return more than just a single image.
2020-05-04 20:01:43 +00:00
pixel_criterion: l1
pixel_weight: !!float 1e-2
feature_criterion: l1
feature_weight: 1
feature_weight_decay: 1
feature_weight_decay_steps: 500
feature_weight_minimum: 1
gan_type: gan # gan | ragan
gan_weight: !!float 1e-2
Random config changes
2020-05-06 23:25:48 +00:00
D_update_ratio: 2
D_init_iters: 0
Add support for passthrough disc/gen Add RRDBNetXL, which performs processing at multiple image sizes. Add DiscResnet_passthrough, which allows passthrough of image at different sizes for discrimination. Adjust the rest of the repo to allow generators that return more than just a single image.
2020-05-04 20:01:43 +00:00
manual_seed: 10
val_freq: !!float 5e2
#### logger
logger:
print_freq: 50
save_checkpoint_freq: !!float 5e2
Reference in New Issue Copy Permalink