Enable HighToLowResNet to do a 1:1 transform

This commit is contained in:
James Betker 2020-04-25 21:36:32 -06:00
parent 35bd1ecae4
commit 02ff4a57fd
2 changed files with 42 additions and 28 deletions

View File

@ -27,14 +27,18 @@ class HighToLowResNet(nn.Module):
self.trunk_lores = arch_util.make_layer(basic_block2, nb - 15)
# downsampling
if self.downscale == 4:
if self.downscale == 4 or self.downscale == 1:
self.downconv1 = nn.Conv2d(nf, nf, 3, stride=2, padding=1, bias=True)
self.downconv2 = nn.Conv2d(nf, nf*2, 3, stride=2, padding=1, bias=True)
else:
raise EnvironmentError("Requested downscale not supported: %i" % (downscale,))
self.HRconv = nn.Conv2d(nf*2, nf*2, 3, stride=1, padding=1, bias=True)
self.conv_last = nn.Conv2d(nf*2, out_nc, 3, stride=1, padding=1, bias=True)
if self.downscale == 4:
self.conv_last = nn.Conv2d(nf*2, out_nc, 3, stride=1, padding=1, bias=True)
else:
self.pixel_shuffle = nn.PixelShuffle(4)
self.conv_last = nn.Conv2d(int(nf/8), out_nc, 3, stride=1, padding=1, bias=True)
# activation function
self.lrelu = nn.LeakyReLU(negative_slope=0.1, inplace=True)
@ -51,13 +55,22 @@ class HighToLowResNet(nn.Module):
out = self.lrelu(self.conv_first(out))
out = self.trunk_hires(out)
if self.downscale == 4:
if self.downscale == 4 or self.downscale == 1:
out = self.lrelu(self.downconv1(out))
out = self.trunk_medres(out)
out = self.lrelu(self.downconv2(out))
out = self.trunk_lores(out)
out = self.conv_last(self.lrelu(self.HRconv(out)))
base = F.interpolate(x, scale_factor=1/self.downscale, mode='bilinear', align_corners=False)
if self.downscale == 1:
out = self.lrelu(self.pixel_shuffle(self.HRconv(out)))
out = self.conv_last(out)
else:
out = self.conv_last(self.lrelu(self.HRconv(out)))
if self.downscale == 1:
base = x
else:
base = F.interpolate(x, scale_factor=1/self.downscale, mode='bilinear', align_corners=False)
out += base
return out

View File

@ -1,9 +1,9 @@
#### general settings
name: downsample_GAN_blacked
name: corruptGAN_4k_lqprn_closeup_hq_to_hq
use_tb_logger: true
model: srgan
distortion: sr
scale: .25
model: corruptgan
distortion: downsample
scale: 1
gpu_ids: [0]
amp_opt_level: O1
@ -11,41 +11,41 @@ amp_opt_level: O1
datasets:
train:
name: blacked
mode: GTLQ
dataroot_GT: ../datasets/blacked/train/hr
dataroot_LQ: ../datasets/lqprn/train/lr
mode: downsample
dataroot_GT: K:\\4k6k\\4k_closeup\\hr
dataroot_LQ: E:\\4k6k\\adrianna\\for_training\\hr
mismatched_Data_OK: true
use_shuffle: true
n_workers: 0 # per GPU
batch_size: 7
target_size: 64
n_workers: 4 # per GPU
batch_size: 16
target_size: 256
use_flip: false
use_rot: false
color: RGB
val:
name: blacked_val
mode: LQGT
dataroot_GT: ../datasets/vrp/validation/hr
dataroot_LQ: ../datasets/vrp/validation/lr
mode: downsample
target_size: 256
dataroot_GT: ../datasets/blacked/val/hr
dataroot_LQ: ../datasets/blacked/val/hr
#### network structures
network_G:
which_model_G: RRDBNet
which_model_G: HighToLowResNet
in_nc: 3
out_nc: 3
nf: 64
nb: 23
nb: 56
network_D:
which_model_D: discriminator_vgg_128
in_nc: 3
nf: 64
nf: 128
#### path
path:
#pretrain_model_G: ../experiments/blacked_gen_20000_epochs.pth
#pretrain_model_D: ../experiments/blacked_disc_20000_epochs.pth
strict_load: true
pretrain_model_G: ~
resume_state: ~
strict_load: true
#### training settings: learning rate scheme, loss
train:
@ -61,16 +61,17 @@ train:
niter: 400000
warmup_iter: -1 # no warm up
lr_steps: [10000, 30000, 50000, 70000]
lr_steps: [4000, 8000, 12000, 15000, 20000]
lr_gamma: 0.5
pixel_criterion: l1
pixel_weight: !!float 1e-2
feature_criterion: l1
feature_weight: 0
gan_type: ragan # gan | ragan
gan_weight: 1
gan_weight: !!float 1e-1
D_update_ratio: 1
D_update_ratio: 2
D_init_iters: 0
manual_seed: 10