Skip to content

Latest commit

 

History

History
88 lines (74 loc) · 3.21 KB

README .md

File metadata and controls

88 lines (74 loc) · 3.21 KB
Raw

Multi-scale super-resolution generation of low-resolution scanned pathological images

This is the implementation of the models and source code for the " Multi-scale super-resolution generation of low-resolution scanned pathological images"。

File description

  • train_adv.py: code to train our generator with loss,train_feature_extractor loss,inlucde configuration for file paths and hyper parameters for networks.
  • predict_adv:code to test our model.
  • rrdn_adv:gans,it can carry out continuous multi-stage magnification, from 2 times to 4 times to 8 times, of which 4 times magnification is the best.
  • weights: weight model trained by train_adv.py

#data processing Since we are using pathology images data, before training, we have to crop the hr data into 128size lr pictures. Of course, you can also use DIK2 data for training.

Usage for training

Create the models

from ISR.models import RRDN
from ISR.models import Discriminator
from ISR.models import Cut_VGG19

lr_train_patch_size = 64
layers_to_extract = [5, 9]
scale = 8
hr_train_patch_size = lr_train_patch_size * scale

rrdn = RRDN_ADV(arch_params={'C':4, 'D':3, 'G':64, 'G0':64, 'T':10, 'x':scale}, patch_size=lr_train_patch_size)
f_ext = Cut_VGG19(patch_size=None, layers_to_extract=layers_to_extract)
discr = Discriminator(patch_size=hr_train_patch_size, kernel_size=3)

Create a Trainer object using the desired settings and give it the models (generand'f_ext and discr are optional)

from ISR.train import trainer_adv
loss_weights = {
  'generator': 0.06,
  'feature_extractor': 0.083,
  'discriminator': 0.01
}
losses = {
  'generator': 'mae',
  'feature_extractor': 'mse',
  'discriminator': 'binary_crossentropy'
}

log_dirs = {'logs': './logs', 'weights': './weights'}

learning_rate = {'initial_value': 0.0004, 'decay_factor': 0.5, 'decay_frequency': 30}

flatness = {'min': 0.0, 'max': 0.15, 'increase': 0.01, 'increase_frequency': 5}

trainer = Trainer_adv(
    generator=rrdn,
    discriminator=discr,
    feature_extractor=f_ext,
    lr_train_dir='./data/lr128_data',
    hr_train_dir=['/data/hr256_data','/data/hr512_data','/data/hr1024_data'],
    lr_valid_dir='./data/lr128_val',
    hr_valid_dir=['/data/hr256_val','/data/hr512_val','/data/hr1024_val'],
   loss_weights=loss_weights,
    learning_rate=learning_rate,
    flatness=flatness,
    dataname='image_dataset',
    log_dirs=log_dirs,
    #weights_generator='rrdn-C4-D3-G64-G064-T10-x8_best-val_generator_PSNR_Y_epoch109.hdf5/',
    weights_generator=None,
    weights_discriminator=None,
    n_validation=40,
)

Start training(epochsand'per_epochs and batch_size are optional)

trainer.train(
    epochs=120,
    steps_per_epoch=500,
    batch_size=2,
    monitored_metrics={'val_PSNR_Y': 'max'}
)

Usage for testing

  • Set test paths for test set and trained models in predict_adv.py (the trained models put in weights folder and must chose the best rrdn models)
  • Run predict_adv.py

Acknowledgement