Image-to-Image
PyTorch
ONNX
TensorRT
English
super-resolution
image-restoration
sisr
onnxruntime
spandrel
chainner
edge
deployment
mit
Instructions to use Phips/PULSE with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- TensorRT
How to use Phips/PULSE with TensorRT:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
Clean release layout
Browse files
configs/2x_pulse_fast_pretrain_pd12m.yml
DELETED
|
@@ -1,88 +0,0 @@
|
|
| 1 |
-
name: 2x_pulse_fast_finetune_full_urban_200k
|
| 2 |
-
scale: 2
|
| 3 |
-
use_amp: true
|
| 4 |
-
amp_bf16: true
|
| 5 |
-
use_channels_last: true
|
| 6 |
-
fast_matmul: false
|
| 7 |
-
use_compile: false
|
| 8 |
-
compile_mode: default
|
| 9 |
-
num_gpu: auto
|
| 10 |
-
manual_seed: 1024
|
| 11 |
-
datasets:
|
| 12 |
-
train:
|
| 13 |
-
name: LUCID Train
|
| 14 |
-
type: pairedimagedataset
|
| 15 |
-
dataroot_gt:
|
| 16 |
-
- /home/phips/Documents/dataset/pd12m/new/pd12m_train_full_gt
|
| 17 |
-
dataroot_lq:
|
| 18 |
-
- /home/phips/Documents/dataset/pd12m/new/pd12m_train_full_lr_x2
|
| 19 |
-
lq_size: 256
|
| 20 |
-
use_hflip: true
|
| 21 |
-
use_rot: true
|
| 22 |
-
num_worker_per_gpu: 8
|
| 23 |
-
batch_size_per_gpu: 8
|
| 24 |
-
accum_iter: 1
|
| 25 |
-
val:
|
| 26 |
-
name: Urban100 x2
|
| 27 |
-
type: pairedimagedataset
|
| 28 |
-
dataroot_gt:
|
| 29 |
-
- /home/phips/Documents/dataset/Urban100/HR
|
| 30 |
-
dataroot_lq:
|
| 31 |
-
- datasets/val/urban100/lr_x2
|
| 32 |
-
filename_tmpl: '{}x2'
|
| 33 |
-
network_g:
|
| 34 |
-
type: pulse_fast
|
| 35 |
-
path:
|
| 36 |
-
pretrain_network_g: experiments/2x_PULSE_fast_pretrain/models/net_g_ema_latest.safetensors
|
| 37 |
-
param_key_g: null
|
| 38 |
-
strict_load_g: true
|
| 39 |
-
resume_state: null
|
| 40 |
-
train:
|
| 41 |
-
ema_decay: 0.999
|
| 42 |
-
ema_power: 0.75
|
| 43 |
-
grad_clip: false
|
| 44 |
-
grad_clip_max_norm: 100
|
| 45 |
-
optim_g:
|
| 46 |
-
type: AdamW
|
| 47 |
-
lr: 0.0001
|
| 48 |
-
weight_decay: 0
|
| 49 |
-
betas:
|
| 50 |
-
- 0.9
|
| 51 |
-
- 0.99
|
| 52 |
-
scheduler:
|
| 53 |
-
type: MultiStepLR
|
| 54 |
-
milestones:
|
| 55 |
-
- 100000
|
| 56 |
-
- 160000
|
| 57 |
-
gamma: 0.5
|
| 58 |
-
total_iter: 200000
|
| 59 |
-
warmup_iter: -1
|
| 60 |
-
losses:
|
| 61 |
-
- type: charbonnierloss
|
| 62 |
-
loss_weight: 1.0
|
| 63 |
-
val:
|
| 64 |
-
val_enabled: true
|
| 65 |
-
val_freq: 5000
|
| 66 |
-
save_img: false
|
| 67 |
-
tile_size: 512
|
| 68 |
-
tile_overlap: 16
|
| 69 |
-
metrics_enabled: true
|
| 70 |
-
metrics:
|
| 71 |
-
psnr:
|
| 72 |
-
type: calculate_psnr
|
| 73 |
-
crop_border: 4
|
| 74 |
-
test_y_channel: true
|
| 75 |
-
ssim:
|
| 76 |
-
type: calculate_ssim
|
| 77 |
-
crop_border: 4
|
| 78 |
-
test_y_channel: true
|
| 79 |
-
topiq:
|
| 80 |
-
type: calculate_topiq
|
| 81 |
-
lpips:
|
| 82 |
-
type: calculate_lpips
|
| 83 |
-
better: lower
|
| 84 |
-
logger:
|
| 85 |
-
print_freq: 500
|
| 86 |
-
save_checkpoint_freq: 10000
|
| 87 |
-
save_checkpoint_format: safetensors
|
| 88 |
-
use_tb_logger: true
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|