Skip to content

Latest commit

 

History

History
15 lines (10 loc) · 3.39 KB

README.md

File metadata and controls

15 lines (10 loc) · 3.39 KB
Raw

Image-based parameter-tuning of a noise filter using Deep Learning

This project explores a new direction in explainable Artificial Intelligence (AI). Most state-of-art deep-learning based noise removal algorithms involve some form of neural networks trained end-to-end (input: noisy image, output: clean image). Such methods are inexplicable "black-boxes" since we do not have a clear understanding of their inner workings and how change in input data distribution can affect their denoising performance. In contrast, this project proposes and validates a CNN-based method that tunes the parameter of a computer vision based (fully explainable) denoising algorithm based only on the noisy input image. Thus, we get the best of both worlds: explain-ability of vision and learning-based regression.

Please cite the below paper if you use the code in its original or modified form:

S. Mukherjee, N. K. Kottayil, X. Sun, and I. Cheng, “CNN-based Real-Time Parameter Tuning for Optimizing Denoising Filter Performance,” in F. Karray, A. Campilho, A. Yu (eds) Image Analysis and Recognition, ICIAR 2019, Lecture Notes in Computer Science, Springer International Publishing, vol 11662, pp. 112–125.

Guidelines

  1. Build the training dataset by running the dataset script create_assorted_dataset_LNET.py, setting train = True for the desired noise_lvl inside the script. The script is written in a way that allows you to source images files with two different extensions from two folders while building the training dataset. This is similar to what we do in our paper.
  2. To use the GPU implementation of BM3D (BM3D-GPU), as in our paper, compile the required binaries. Then, set bm3d_cpu = False inside the dataset script and set the bm3dobj_loc path to your compiled BM3D-GPU binary path.
  3. The model script LNet.py has the implementation of the CNN we use to predict the λ3D parameter in our paper.
  4. Run the training script learn_predict_L3d_noisysrc.py by similarly setting the noise_lvl and bm3d_cpu = False as before.
  5. If you want to evaluate the trained model on a test set of images, you can create your test set by running the dataset script create_assorted_dataset_LNET.py, after setting train = False for the desired noise_lvl inside the script, along with bm3d_cpu = False and the bm3dobj_loc path as before. Then, you can run the evaluation script eval_L3d_noisysrc.py twice, first to generate the predicted λ3D values for the test images, and then to evaluate BM3D performance in terms of average MSE and SSIM scores on the test set using the predicted λ3D values, by changing Line 76 to if False:.