Dacon AISR

🔥 Team : 모두연 포레버
🔥 Member : 박수철, 장진우, 윤성국, 양성모

Descriptions

Competition Link : AI 양재 허브 인공지능 오픈소스 경진대회

[주제]

이미지 초해상화(Image Sper-Resolution)를 위한 AI 알고리즘 개발

[목적 및 배경]

인공지능 오픈소스를 활용하여 이미지 초해상화 문제를 해결함으로써 오픈소스 생태계와 컴퓨터 비전 분야에 기여

[설명]

품질이 저하된 저해상도 촬영 이미지(512x512)를 고품질의 고해상도 촬영 이미지(2048x2048)로 생성

[평가 산식]

PSNR(Peak Signal-to-Noise Ratio) $$PSNR = 10log_{10}(\frac{R^2}{MSE})$$

🚩 Score

Private Score
▶️ 25.00327 [1st]
Public Score
▶️ 25.96657 [2nd]

💻 Environment

<Main>
OS : Ubuntu 20.04.5 LTS
CPU : Intel(R) Xeon(R) Gold 6230R CPU @ 2.10GHz
GPU : A5000 * 4

&

<Sub>
Colab Pro+

✅ Competition strategies

1. Data Augmentation

Rotation
Flip

2. Post Processing

Geometric Self-Ensemble [https://arxiv.org/pdf/1707.02921.pdf]

3. Loss function

MSELoss ( PSNR socre is directly related with MSE )

4. gt_size increase gt_size = 64 -> 512

5. Multistep LR optimizer

Decays the learning rate of each parameter group by gamma once the number of epoch reaches one of the milestones of 200,000
In the second half of the learning, detailed learning is performed.

Main configuration & Hyper parameters

manuel_seed : 0
gt_size : 512
num_feat : 64

train:
  ema_decay: 0.999
  optim_g:
    type: Adam
    lr: !!float 1e-4
    weight_decay: 0
    betas: [0.9, 0.99]

pixel_opt:
  type: MSELoss
  loss_weight: 1.0
  reduction: mean

Configuration in ./options/finetune_realesrnet_x4plus_pairdata.yml

# general settings
name: finetune_RealESRNetx4plus_400k_pairdata
model_type: RealESRNetModel
scale: 4
num_gpu: auto
manual_seed: 0

# USM the ground-truth
l1_gt_usm: True
percep_gt_usm: True
gan_gt_usm: False

high_order_degradation: False # do not use the high-order degradation generation process

# dataset and data loader settings
datasets:

  train: # the 1st test dataset
    name: Dacon_train
    type: PairedImageDataset
    dataroot_gt: ./inputs/train/hr
    dataroot_lq: ./inputs/train/lr
    io_backend:
      type: disk

    gt_size: 512
    use_hflip: true
    use_rot: true

    # data loader
    use_shuffle: true
    num_worker_per_gpu: 4
    batch_size_per_gpu: 4
    dataset_enlarge_ratio: 1
    prefetch_mode: ~

# network structures
network_g:
  type: RRDBNet
  num_in_ch: 3
  num_out_ch: 3
  num_feat: 96
  num_block: 23
  num_grow_ch: 32

network_d:
  type: UNetDiscriminatorSN
  num_in_ch: 3
  num_feat: 96
  skip_connection: True

# path
path:
  # use the pre-trained Real-ESRNet model
  #pretrain_network_g: experiments/pretrained_models/RealESRGAN_x4plus.pth
  pretrain_network_g: ./weights/net_g_905000.pth
  param_key_g: params_ema
  strict_load_g: false
  resume_state: ~

# training settings
train:
  ema_decay: 0.999
  optim_g:
    type: Adam
    lr: !!float 1e-4
    weight_decay: 0
    betas: [0.9, 0.99]

  scheduler:
    type: MultiStepLR
    milestones: [400000]
    gamma: 0.5

  total_iter: 999999999
  warmup_iter: -1  # no warm up

  # losses
  pixel_opt:
    type: MSELoss
    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

🔨 Installation

Clone repo

git clone https://github.com/Jinwoo1126/Dacon_SR.git
cd Dacon_SR

Install dependent packages

# Install basicsr - https://github.com/xinntao/BasicSR
# We use BasicSR for both training and inference
pip install basicsr
# facexlib and gfpgan are for face enhancement
pip install facexlib
pip install gfpgan
pip install -r requirements.txt
python setup.py develop

🏃 Running Code

Pretrained Model

RRDBNet_x4_Pretrained_Model

File Structure

./Dacon_SR
├── inputs
│   ├── train
|   |   ├── hr
│   │   |   ├── 0000.png
│   │   |   ├── 0001.png
│   │   |   ├── 0002.png
|   |   |    ...
│   │   ├── lr
|   |       ├── 0000.png
|   |       ├── 0001.png
|   |       ├── 0002.png
|   |        ...
│   ├── test
│   │   ├── 20000.png
│   │   ├── 20001.png
│   │   ├── ...
├── tags
├── ...
├── weight
│   ├── net_g_905000.pth

Inference

python inference_rrdbnetrot.py --model_path=./weights/net_g_905000.pth --input=./inputs/test/lr --suffix=''

Results are in the results folder

(Optional) Finetuning with pretrained model

python realesrgan/train.py -opt options/finetune_realesrnet_x4plus_pairdata.yml

experiments result : ./experiments/[your experiments name]

(default='finetune_RealESRNetx4plus_400k_pairdata')

The test result can be slightly diff from ours due to the different hardware architectures.

reproducible issues

https://discuss.pytorch.org/t/different-training-results-on-different-machines-with-simplified-test-code/59378/11

https://discuss.pytorch.org/t/different-result-on-different-gpu/102502

kooock/Dacon_SR