"vscode:/vscode.git/clone" did not exist on "2b82c5a8a1bc783632e10f87d9e923f5ab302085"
Commit 2ac5586e authored by Rayyyyy's avatar Rayyyyy
Browse files

first commit

parents
Pipeline #784 canceled with stages
name: HAT_SRx3_ImageNet-pretrain
model_type: HATModel
scale: 3
num_gpu: 1 # set num_gpu: 0 for cpu mode
manual_seed: 0
datasets:
test_1: # the 1st test dataset
name: Set5
type: PairedImageDataset
dataroot_gt: ./datasets/Set5/GTmod3
dataroot_lq: ./datasets/Set5/LRbicx3
io_backend:
type: disk
# test_2: # the 2nd test dataset
# name: Set14
# type: PairedImageDataset
# dataroot_gt: ./datasets/Set14/GTmod3
# dataroot_lq: ./datasets/Set14/LRbicx3
# io_backend:
# type: disk
# test_3:
# name: Urban100
# type: PairedImageDataset
# dataroot_gt: ./datasets/urban100/GTmod3
# dataroot_lq: ./datasets/urban100/LRbicx3
# io_backend:
# type: disk
# test_4:
# name: BSDS100
# type: PairedImageDataset
# dataroot_gt: ./datasets/BSDS100/GTmod3
# dataroot_lq: ./datasets/BSDS100/LRbicx3
# io_backend:
# type: disk
# test_5:
# name: Manga109
# type: PairedImageDataset
# dataroot_gt: ./datasets/manga109/GTmod3
# dataroot_lq: ./datasets/manga109/LRbicx3
# io_backend:
# type: disk
# network structures
network_g:
type: HAT
upscale: 3
in_chans: 3
img_size: 64
window_size: 16
compress_ratio: 3
squeeze_factor: 30
conv_scale: 0.01
overlap_ratio: 0.5
img_range: 1.
depths: [6, 6, 6, 6, 6, 6]
embed_dim: 180
num_heads: [6, 6, 6, 6, 6, 6]
mlp_ratio: 2
upsampler: 'pixelshuffle'
resi_connection: '1conv'
# path
path:
pretrain_network_g: ./experiments/pretrained_models/HAT_SRx3_ImageNet-pretrain.pth
strict_load_g: true
param_key_g: 'params_ema'
# validation settings
val:
save_img: true
suffix: ~ # add suffix to saved images, if None, use exp name
metrics:
psnr: # metric name, can be arbitrary
type: calculate_psnr
crop_border: 3
test_y_channel: true
ssim:
type: calculate_ssim
crop_border: 3
test_y_channel: true
name: HAT_SRx4
model_type: HATModel
scale: 4
num_gpu: 1 # set num_gpu: 0 for cpu mode
manual_seed: 0
datasets:
test_1: # the 1st test dataset
name: Set5
type: PairedImageDataset
dataroot_gt: ./datasets/Set5/GTmod4
dataroot_lq: ./datasets/Set5/LRbicx4
io_backend:
type: disk
# test_2: # the 2nd test dataset
# name: Set14
# type: PairedImageDataset
# dataroot_gt: ./datasets/Set14/GTmod4
# dataroot_lq: ./datasets/Set14/LRbicx4
# io_backend:
# type: disk
# test_3:
# name: Urban100
# type: PairedImageDataset
# dataroot_gt: ./datasets/urban100/GTmod4
# dataroot_lq: ./datasets/urban100/LRbicx4
# io_backend:
# type: disk
# test_4:
# name: BSDS100
# type: PairedImageDataset
# dataroot_gt: ./datasets/BSDS100/GTmod4
# dataroot_lq: ./datasets/BSDS100/LRbicx4
# io_backend:
# type: disk
# test_5:
# name: Manga109
# type: PairedImageDataset
# dataroot_gt: ./datasets/manga109/GTmod4
# dataroot_lq: ./datasets/manga109/LRbicx4
# io_backend:
# type: disk
# network structures
network_g:
type: HAT
upscale: 4
in_chans: 3
img_size: 64
window_size: 16
compress_ratio: 3
squeeze_factor: 30
conv_scale: 0.01
overlap_ratio: 0.5
img_range: 1.
depths: [6, 6, 6, 6, 6, 6]
embed_dim: 180
num_heads: [6, 6, 6, 6, 6, 6]
mlp_ratio: 2
upsampler: 'pixelshuffle'
resi_connection: '1conv'
# path
path:
pretrain_network_g: ./experiments/pretrained_models/HAT_SRx4.pth
strict_load_g: true
param_key_g: 'params_ema'
# validation settings
val:
save_img: true
suffix: ~ # add suffix to saved images, if None, use exp name
metrics:
psnr: # metric name, can be arbitrary
type: calculate_psnr
crop_border: 4
test_y_channel: true
ssim:
type: calculate_ssim
crop_border: 4
test_y_channel: true
name: HAT_SRx4_ImageNet-LR
model_type: HATModel
scale: 4
num_gpu: 1 # set num_gpu: 0 for cpu mode
manual_seed: 0
tile:
tile_size: 512 # max patch size for the tile mode
tile_pad: 32
datasets:
test_1: # the 1st test dataset
name: custom
type: SingleImageDataset
dataroot_lq: ./datasets/Set5/LRbicx4
io_backend:
type: disk
# network structures
network_g:
type: HAT
upscale: 4
in_chans: 3
img_size: 64
window_size: 16
compress_ratio: 3
squeeze_factor: 30
conv_scale: 0.01
overlap_ratio: 0.5
img_range: 1.
depths: [6, 6, 6, 6, 6, 6]
embed_dim: 180
num_heads: [6, 6, 6, 6, 6, 6]
mlp_ratio: 2
upsampler: 'pixelshuffle'
resi_connection: '1conv'
# path
path:
pretrain_network_g: experiments/pretrained_models/HAT_SRx4_ImageNet-pretrain.pth
strict_load_g: true
param_key_g: 'params_ema'
# validation settings
val:
save_img: true
suffix: ~ # add suffix to saved images, if None, use exp name
name: HAT_SRx4_ImageNet-pretrain
model_type: HATModel
scale: 4
num_gpu: 1 # set num_gpu: 0 for cpu mode
manual_seed: 0
datasets:
test_1: # the 1st test dataset
name: Set5
type: PairedImageDataset
dataroot_gt: ./datasets/Set5/GTmod12
dataroot_lq: ./datasets/Set5/LRbicx4
io_backend:
type: disk
# test_2: # the 2nd test dataset
# name: Set14
# type: PairedImageDataset
# dataroot_gt: ./datasets/Set14/GTmod12
# dataroot_lq: ./datasets/Set14/LRbicx4
# io_backend:
# type: disk
# test_3:
# name: Urban100
# type: PairedImageDataset
# dataroot_gt: ./datasets/urban100/GTmod4
# dataroot_lq: ./datasets/urban100/LRbicx4
# io_backend:
# type: disk
# test_4:
# name: BSDS100
# type: PairedImageDataset
# dataroot_gt: ./datasets/BSDS100/GTmod4
# dataroot_lq: ./datasets/BSDS100/LRbicx4
# io_backend:
# type: disk
# test_5:
# name: Manga109
# type: PairedImageDataset
# dataroot_gt: ./datasets/manga109/GTmod4
# dataroot_lq: ./datasets/manga109/LRbicx4
# io_backend:
# type: disk
# network structures
network_g:
type: HAT
upscale: 4
in_chans: 3
img_size: 64
window_size: 16
compress_ratio: 3
squeeze_factor: 30
conv_scale: 0.01
overlap_ratio: 0.5
img_range: 1.
depths: [6, 6, 6, 6, 6, 6]
embed_dim: 180
num_heads: [6, 6, 6, 6, 6, 6]
mlp_ratio: 2
upsampler: 'pixelshuffle'
resi_connection: '1conv'
# path
path:
pretrain_network_g: ./experiments/pretrained_models/HAT_SRx4_ImageNet-pretrain.pth
strict_load_g: true
param_key_g: 'params_ema'
# validation settings
val:
save_img: true
suffix: ~ # add suffix to saved images, if None, use exp name
metrics:
psnr: # metric name, can be arbitrary
type: calculate_psnr
crop_border: 4
test_y_channel: true
ssim:
type: calculate_ssim
crop_border: 4
test_y_channel: true
name: HAT_SRx4_ImageNet-pretrain
model_type: HATModel
scale: 4
num_gpu: 1 # set num_gpu: 0 for cpu mode
manual_seed: 0
tile: # use the tile mode for limited GPU memory when testing.
tile_size: 256 # the higher, the more utilized GPU memory and the less performance change against the full image. must be an integer multiple of the window size.
tile_pad: 32 # overlapping between adjacency patches.must be an integer multiple of the window size.
datasets:
test_1: # the 1st test dataset
name: Set5
type: PairedImageDataset
dataroot_gt: ./datasets/Set5/GTmod4
dataroot_lq: ./datasets/Set5/LRbicx4
io_backend:
type: disk
# test_2: # the 2nd test dataset
# name: Set14
# type: PairedImageDataset
# dataroot_gt: ./datasets/Set14/GTmod4
# dataroot_lq: ./datasets/Set14/LRbicx4
# io_backend:
# type: disk
test_3:
name: Urban100
type: PairedImageDataset
dataroot_gt: ./datasets/urban100/GTmod4
dataroot_lq: ./datasets/urban100/LRbicx4
io_backend:
type: disk
# test_4:
# name: BSDS100
# type: PairedImageDataset
# dataroot_gt: ./datasets/BSDS100/GTmod4
# dataroot_lq: ./datasets/BSDS100/LRbicx4
# io_backend:
# type: disk
# test_5:
# name: Manga109
# type: PairedImageDataset
# dataroot_gt: ./datasets/manga109/GTmod4
# dataroot_lq: ./datasets/manga109/LRbicx4
# io_backend:
# type: disk
# network structures
network_g:
type: HAT
upscale: 4
in_chans: 3
img_size: 64
window_size: 16
compress_ratio: 3
squeeze_factor: 30
conv_scale: 0.01
overlap_ratio: 0.5
img_range: 1.
depths: [6, 6, 6, 6, 6, 6]
embed_dim: 180
num_heads: [6, 6, 6, 6, 6, 6]
mlp_ratio: 2
upsampler: 'pixelshuffle'
resi_connection: '1conv'
# path
path:
pretrain_network_g: ./experiments/pretrained_models/HAT_SRx4_ImageNet-pretrain.pth
strict_load_g: true
param_key_g: 'params_ema'
# validation settings
val:
save_img: true
suffix: ~ # add suffix to saved images, if None, use exp name
metrics:
psnr: # metric name, can be arbitrary
type: calculate_psnr
crop_border: 4
test_y_channel: true
ssim:
type: calculate_ssim
crop_border: 4
test_y_channel: true
# general settings
name: train_HAT-L_SRx2_ImageNet_from_scratch
model_type: HATModel
scale: 2
num_gpu: auto
manual_seed: 0
# dataset and data loader settings
datasets:
train:
name: ImageNet
type: ImageNetPairedDataset
dataroot_gt: datasets/ImageNet/GT
meta_info_file: hat/data/meta_info/meta_info_ImageNet_GT.txt
io_backend:
type: disk
gt_size: 128
use_hflip: true
use_rot: true
# data loader
use_shuffle: true
num_worker_per_gpu: 6
batch_size_per_gpu: 4
dataset_enlarge_ratio: 1
prefetch_mode: ~
val_1:
name: Set5
type: PairedImageDataset
dataroot_gt: ./datasets/Set5/GTmod2
dataroot_lq: ./datasets/Set5/LRbicx2
io_backend:
type: disk
val_2:
name: Set14
type: PairedImageDataset
dataroot_gt: ./datasets/Set14/GTmod2
dataroot_lq: ./datasets/Set14/LRbicx2
io_backend:
type: disk
# val_3:
# name: Urban100
# type: PairedImageDataset
# dataroot_gt: ./datasets/urban100/GTmod2
# dataroot_lq: ./datasets/urban100/LRbicx2
# io_backend:
# type: disk
# network structures
network_g:
type: HAT
upscale: 2
in_chans: 3
img_size: 64
window_size: 16
compress_ratio: 3
squeeze_factor: 30
conv_scale: 0.01
overlap_ratio: 0.5
img_range: 1.
depths: [6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6]
embed_dim: 180
num_heads: [6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6]
mlp_ratio: 2
upsampler: 'pixelshuffle'
resi_connection: '1conv'
# path
path:
pretrain_network_g: ~
strict_load_g: true
resume_state: ~
# training settings
train:
ema_decay: 0.999
optim_g:
type: Adam
lr: !!float 2e-4
weight_decay: 0
betas: [0.9, 0.99]
scheduler:
type: MultiStepLR
milestones: [300000, 500000, 650000, 700000, 750000]
gamma: 0.5
total_iter: 800000
warmup_iter: -1 # no warm up
# losses
pixel_opt:
type: L1Loss
loss_weight: 1.0
reduction: mean
# validation settings
val:
val_freq: !!float 1e4
save_img: false
pbar: False
metrics:
psnr:
type: calculate_psnr
crop_border: 2
test_y_channel: true
better: higher # the higher, the better. Default: higher
ssim:
type: calculate_ssim
crop_border: 2
test_y_channel: true
better: higher # the higher, the better. Default: higher
# logging settings
logger:
print_freq: 100
save_checkpoint_freq: !!float 1e4
use_tb_logger: true
wandb:
project: ~
resume_id: ~
# dist training settings
dist_params:
backend: nccl
port: 29500
# general settings
name: train_HAT-L_SRx2_finetune_from_ImageNet_pretrain
model_type: HATModel
scale: 2
num_gpu: auto
manual_seed: 0
# dataset and data loader settings
datasets:
train:
name: DF2K
type: PairedImageDataset
dataroot_gt: datasets/DF2K/DF2K_HR_sub
dataroot_lq: datasets/DF2K/DF2K_bicx2_sub
meta_info_file: hat/data/meta_info/meta_info_DF2Ksub_GT.txt
io_backend:
type: disk
gt_size: 128
use_hflip: true
use_rot: true
# data loader
use_shuffle: true
num_worker_per_gpu: 6
batch_size_per_gpu: 4
dataset_enlarge_ratio: 1
prefetch_mode: ~
val_1:
name: Set5
type: PairedImageDataset
dataroot_gt: ./datasets/Set5/GTmod2
dataroot_lq: ./datasets/Set5/LRbicx2
io_backend:
type: disk
val_2:
name: Set14
type: PairedImageDataset
dataroot_gt: ./datasets/Set14/GTmod2
dataroot_lq: ./datasets/Set14/LRbicx2
io_backend:
type: disk
# val_3:
# name: Urban100
# type: PairedImageDataset
# dataroot_gt: ./datasets/urban100/GTmod2
# dataroot_lq: ./datasets/urban100/LRbicx2
# io_backend:
# type: disk
# network structures
network_g:
type: HAT
upscale: 2
in_chans: 3
img_size: 64
window_size: 16
compress_ratio: 3
squeeze_factor: 30
conv_scale: 0.01
overlap_ratio: 0.5
img_range: 1.
depths: [6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6]
embed_dim: 180
num_heads: [6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6]
mlp_ratio: 2
upsampler: 'pixelshuffle'
resi_connection: '1conv'
# path
path:
pretrain_network_g: ./experiments/train_HAT-L_SRx2_ImageNet_from_scratch/models/net_g_latest.pth
param_key_g: 'params_ema'
strict_load_g: true
resume_state: ~
# training settings
train:
ema_decay: 0.999
optim_g:
type: Adam
lr: !!float 1e-5
weight_decay: 0
betas: [0.9, 0.99]
scheduler:
type: MultiStepLR
milestones: [125000, 200000, 225000, 240000]
gamma: 0.5
total_iter: 250000
warmup_iter: -1 # no warm up
# losses
pixel_opt:
type: L1Loss
loss_weight: 1.0
reduction: mean
# validation settings
val:
val_freq: !!float 5e3
save_img: false
pbar: False
metrics:
psnr:
type: calculate_psnr
crop_border: 2
test_y_channel: true
better: higher # the higher, the better. Default: higher
ssim:
type: calculate_ssim
crop_border: 2
test_y_channel: true
better: higher # the higher, the better. Default: higher
# logging settings
logger:
print_freq: 100
save_checkpoint_freq: !!float 5e3
use_tb_logger: true
wandb:
project: ~
resume_id: ~
# dist training settings
dist_params:
backend: nccl
port: 29500
# general settings
name: train_HAT-L_SRx3_ImageNet_from_scratch
model_type: HATModel
scale: 3
num_gpu: auto
manual_seed: 0
# dataset and data loader settings
datasets:
train:
name: ImageNet
type: ImageNetPairedDataset
dataroot_gt: datasets/ImageNet/GT
meta_info_file: hat/data/meta_info/meta_info_ImageNet_GT.txt
io_backend:
type: disk
gt_size: 192
use_hflip: true
use_rot: true
# data loader
use_shuffle: true
num_worker_per_gpu: 6
batch_size_per_gpu: 4
dataset_enlarge_ratio: 1
prefetch_mode: ~
val_1:
name: Set5
type: PairedImageDataset
dataroot_gt: ./datasets/Set5/GTmod3
dataroot_lq: ./datasets/Set5/LRbicx3
io_backend:
type: disk
val_2:
name: Set14
type: PairedImageDataset
dataroot_gt: ./datasets/Set14/GTmod3
dataroot_lq: ./datasets/Set14/LRbicx3
io_backend:
type: disk
# val_3:
# name: Urban100
# type: PairedImageDataset
# dataroot_gt: ./datasets/urban100/GTmod3
# dataroot_lq: ./datasets/urban100/LRbicx3
# io_backend:
# type: disk
# network structures
network_g:
type: HAT
upscale: 3
in_chans: 3
img_size: 64
window_size: 16
compress_ratio: 3
squeeze_factor: 30
conv_scale: 0.01
overlap_ratio: 0.5
img_range: 1.
depths: [6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6]
embed_dim: 180
num_heads: [6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6]
mlp_ratio: 2
upsampler: 'pixelshuffle'
resi_connection: '1conv'
# path
path:
pretrain_network_g: ~
strict_load_g: true
resume_state: ~
# training settings
train:
ema_decay: 0.999
optim_g:
type: Adam
lr: !!float 2e-4
weight_decay: 0
betas: [0.9, 0.99]
scheduler:
type: MultiStepLR
milestones: [300000, 500000, 650000, 700000, 750000]
gamma: 0.5
total_iter: 800000
warmup_iter: -1 # no warm up
# losses
pixel_opt:
type: L1Loss
loss_weight: 1.0
reduction: mean
# validation settings
val:
val_freq: !!float 1e4
save_img: false
pbar: False
metrics:
psnr:
type: calculate_psnr
crop_border: 3
test_y_channel: true
better: higher # the higher, the better. Default: higher
ssim:
type: calculate_ssim
crop_border: 3
test_y_channel: true
better: higher # the higher, the better. Default: higher
# logging settings
logger:
print_freq: 100
save_checkpoint_freq: !!float 1e4
use_tb_logger: true
wandb:
project: ~
resume_id: ~
# dist training settings
dist_params:
backend: nccl
port: 29500
# general settings
name: train_HAT-L_SRx3_finetune_from_ImageNet_pretrain
model_type: HATModel
scale: 3
num_gpu: auto
manual_seed: 0
# dataset and data loader settings
datasets:
train:
name: DF2K
type: PairedImageDataset
dataroot_gt: datasets/DF2K/DF2K_HR_sub
dataroot_lq: datasets/DF2K/DF2K_bicx3_sub
meta_info_file: hat/data/meta_info/meta_info_DF2Ksub_GT.txt
io_backend:
type: disk
gt_size: 192
use_hflip: true
use_rot: true
# data loader
use_shuffle: true
num_worker_per_gpu: 6
batch_size_per_gpu: 4
dataset_enlarge_ratio: 1
prefetch_mode: ~
val_1:
name: Set5
type: PairedImageDataset
dataroot_gt: ./datasets/Set5/GTmod3
dataroot_lq: ./datasets/Set5/LRbicx3
io_backend:
type: disk
val_2:
name: Set14
type: PairedImageDataset
dataroot_gt: ./datasets/Set14/GTmod3
dataroot_lq: ./datasets/Set14/LRbicx3
io_backend:
type: disk
# val_3:
# name: Urban100
# type: PairedImageDataset
# dataroot_gt: ./datasets/urban100/GTmod3
# dataroot_lq: ./datasets/urban100/LRbicx3
# io_backend:
# type: disk
# network structures
network_g:
type: HAT
upscale: 3
in_chans: 3
img_size: 64
window_size: 16
compress_ratio: 3
squeeze_factor: 30
conv_scale: 0.01
overlap_ratio: 0.5
img_range: 1.
depths: [6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6]
embed_dim: 180
num_heads: [6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6]
mlp_ratio: 2
upsampler: 'pixelshuffle'
resi_connection: '1conv'
# path
path:
pretrain_network_g: ./experiments/train_HAT-L_SRx3_ImageNet_from_scratch/models/net_g_latest.pth
param_key_g: 'params_ema'
strict_load_g: true
resume_state: ~
# training settings
train:
ema_decay: 0.999
optim_g:
type: Adam
lr: !!float 1e-5
weight_decay: 0
betas: [0.9, 0.99]
scheduler:
type: MultiStepLR
milestones: [125000, 200000, 225000, 240000]
gamma: 0.5
total_iter: 250000
warmup_iter: -1 # no warm up
# losses
pixel_opt:
type: L1Loss
loss_weight: 1.0
reduction: mean
# validation settings
val:
val_freq: !!float 5e3
save_img: false
pbar: False
metrics:
psnr:
type: calculate_psnr
crop_border: 3
test_y_channel: true
better: higher # the higher, the better. Default: higher
ssim:
type: calculate_ssim
crop_border: 3
test_y_channel: true
better: higher # the higher, the better. Default: higher
# logging settings
logger:
print_freq: 100
save_checkpoint_freq: !!float 5e3
use_tb_logger: true
wandb:
project: ~
resume_id: ~
# dist training settings
dist_params:
backend: nccl
port: 29500
# general settings
name: train_HAT-L_SRx4_ImageNet_from_scratch
model_type: HATModel
scale: 4
num_gpu: auto
manual_seed: 0
# dataset and data loader settings
datasets:
train:
name: ImageNet
type: ImageNetPairedDataset
dataroot_gt: datasets/ImageNet/GT
meta_info_file: hat/data/meta_info/meta_info_ImageNet_GT.txt
io_backend:
type: disk
gt_size: 256
use_hflip: true
use_rot: true
# data loader
use_shuffle: true
num_worker_per_gpu: 6
batch_size_per_gpu: 4
dataset_enlarge_ratio: 1
prefetch_mode: ~
val_1:
name: Set5
type: PairedImageDataset
dataroot_gt: ./datasets/Set5/GTmod4
dataroot_lq: ./datasets/Set5/LRbicx4
io_backend:
type: disk
val_2:
name: Set14
type: PairedImageDataset
dataroot_gt: ./datasets/Set14/GTmod4
dataroot_lq: ./datasets/Set14/LRbicx4
io_backend:
type: disk
# val_3:
# name: Urban100
# type: PairedImageDataset
# dataroot_gt: ./datasets/urban100/GTmod4
# dataroot_lq: ./datasets/urban100/LRbicx4
# io_backend:
# type: disk
# network structures
network_g:
type: HAT
upscale: 4
in_chans: 3
img_size: 64
window_size: 16
compress_ratio: 3
squeeze_factor: 30
conv_scale: 0.01
overlap_ratio: 0.5
img_range: 1.
depths: [6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6]
embed_dim: 180
num_heads: [6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6]
mlp_ratio: 2
upsampler: 'pixelshuffle'
resi_connection: '1conv'
# path
path:
pretrain_network_g: ~
strict_load_g: true
resume_state: ~
# training settings
train:
ema_decay: 0.999
optim_g:
type: Adam
lr: !!float 2e-4
weight_decay: 0
betas: [0.9, 0.99]
scheduler:
type: MultiStepLR
milestones: [300000, 500000, 650000, 700000, 750000]
gamma: 0.5
total_iter: 800000
warmup_iter: -1 # no warm up
# losses
pixel_opt:
type: L1Loss
loss_weight: 1.0
reduction: mean
# validation settings
val:
val_freq: !!float 1e4
save_img: false
pbar: False
metrics:
psnr:
type: calculate_psnr
crop_border: 4
test_y_channel: true
better: higher # the higher, the better. Default: higher
ssim:
type: calculate_ssim
crop_border: 4
test_y_channel: true
better: higher # the higher, the better. Default: higher
# logging settings
logger:
print_freq: 100
save_checkpoint_freq: !!float 1e4
use_tb_logger: true
wandb:
project: ~
resume_id: ~
# dist training settings
dist_params:
backend: nccl
port: 29500
# general settings
name: train_HAT-L_SRx4_finetune_from_ImageNet_pretrain
model_type: HATModel
scale: 4
num_gpu: auto
manual_seed: 0
# dataset and data loader settings
datasets:
train:
name: DF2K
type: PairedImageDataset
dataroot_gt: datasets/DF2K/DF2K_HR_sub
dataroot_lq: datasets/DF2K/DF2K_bicx4_sub
meta_info_file: hat/data/meta_info/meta_info_DF2Ksub_GT.txt
io_backend:
type: disk
gt_size: 256
use_hflip: true
use_rot: true
# data loader
use_shuffle: true
num_worker_per_gpu: 6
batch_size_per_gpu: 4
dataset_enlarge_ratio: 1
prefetch_mode: ~
val_1:
name: Set5
type: PairedImageDataset
dataroot_gt: ./datasets/Set5/GTmod12
dataroot_lq: ./datasets/Set5/LRbicx4
io_backend:
type: disk
val_2:
name: Set14
type: PairedImageDataset
dataroot_gt: ./datasets/Set14/GTmod12
dataroot_lq: ./datasets/Set14/LRbicx4
io_backend:
type: disk
# val_3:
# name: Urban100
# type: PairedImageDataset
# dataroot_gt: ./datasets/urban100/GTmod12
# dataroot_lq: ./datasets/urban100/LRbicx4
# io_backend:
# type: disk
# network structures
network_g:
type: HAT
upscale: 4
in_chans: 3
img_size: 64
window_size: 16
compress_ratio: 3
squeeze_factor: 30
conv_scale: 0.01
overlap_ratio: 0.5
img_range: 1.
depths: [6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6]
embed_dim: 180
num_heads: [6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6]
mlp_ratio: 2
upsampler: 'pixelshuffle'
resi_connection: '1conv'
# path
path:
pretrain_network_g: ./experiments/train_HAT-L_SRx4_ImageNet_from_scratch/models/net_g_latest.pth
param_key_g: 'params_ema'
strict_load_g: true
resume_state: ~
# training settings
train:
ema_decay: 0.999
optim_g:
type: Adam
lr: !!float 1e-5
weight_decay: 0
betas: [0.9, 0.99]
scheduler:
type: MultiStepLR
milestones: [125000, 200000, 225000, 240000]
gamma: 0.5
total_iter: 250000
warmup_iter: -1 # no warm up
# losses
pixel_opt:
type: L1Loss
loss_weight: 1.0
reduction: mean
# validation settings
val:
val_freq: !!float 5e3
save_img: false
pbar: False
metrics:
psnr:
type: calculate_psnr
crop_border: 4
test_y_channel: true
better: higher # the higher, the better. Default: higher
ssim:
type: calculate_ssim
crop_border: 4
test_y_channel: true
better: higher # the higher, the better. Default: higher
# logging settings
logger:
print_freq: 100
save_checkpoint_freq: !!float 5e3
use_tb_logger: true
wandb:
project: ~
resume_id: ~
# dist training settings
dist_params:
backend: nccl
port: 29500
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
This diff is collapsed.
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment