berikaa/DCASR

DCASR

The DCASR model, trained with the DIV2K dataset, is implemented using TensorFlow. It is capable of image enhancement at x2, x3 and x4 scales.

DCASR model has been proven to give better results than other SR models compared to other SR models. You can find all the details of the model in the source below.

• A novel model for higher performance object detection with deep channel attention super resolution

IMPORTANT: If you are going to use the DCASR model in your academic studies, you must cite the original article.

Model weights saved for x2, x3 and x4:

• DCASR_x2
• DCASR_x3
• DCASR_x4

The model is built in tensorflow 2.14.0.

  pip install tensorflow==2.14.0

Usage

  from tensorflow.keras.models import load_model
  from PIL import Image
  from skimage import io
  import tensorflow as tf
  
  device = "/GPU:0" if tf.config.list_physical_devices('GPU') else "/CPU:0"
  print(f"used device: {device}")
  
  model = load_model("DCASR_x2.keras")
  original = io.imread("input/lr_image.png")
  preds = model.predict_step(original)
  preds_np = preds.numpy()
  predicted_image = Image.fromarray(preds_np.astype('uint8'))
  predicted_image.save("output/hr_image.png")