Image Processing

Image Exposure Correction: CLAHE & ECM

Two implementation for exposure adjustment:

• CLAHE (./exposure/CLAHE/clahe.py)
• ECM model (./exposure/ECM/ECM)

To execute the CLAHE, open your terminal and run:

python clahe.py

Remember to change the data path in the script before execution!

To test the ECM, open your terminal and run:

python test.py --ntest 1000 --results_dir "dataset/ECM_output" --checkpoints_dir checkpoints --name exposure_correction_experiment --loadSize 640 --fineSize 640 --dataroot "dataset" --dir_A input --no_flip --label_nc 3 --how_many 1000 --phase_test_type test_all --which_epoch 100 --netG global --ngf 64 --n_downsample_global 4 --n_blocks_global 9 --batchSize 1

To evaluate the ECM output, run:

python evaluate_new.py --images_path1 dataset/ground_truth --images_path2 dataset/ECM_output/exposure_correction_experiment/test_100/images --metric SSIM

Here is how we finetune the pretrained model with our own images in dataset/train:

python train.py \
  --isTrain True \
  --continue_train True \
  --name exposure_correction_experiment \
  --checkpoints_dir checkpoints \
  --which_epoch 100 \
  --dataroot "my data path" \
  --dir_A train \
  --dir_B ground_truth \
  --loadSize 640 \
  --fineSize 640 \
  --nThreads 0 \
  --gpu_ids 0 \
  --batchSize 1 \
  --niter 50 \
  --l1_image_loss True \
  --perceptual_loss True \
  --label_nc 3 \
  --output_nc 3 \
  --netG global \
  --ngf 64 \
  --n_downsample_global 4 \

There is also a ECM/ECM/README.md from the ECM github repo I cite, with the link to download the pretrained model I used.

Image Deblurring: Wiener Filter & MPRNet

This repository contains two implementations for image deblurring:

• Wiener Filter baseline (./deblur/wiener/wiener.py)
• MPRNet deep learning model (./deblur/mprnet/demo_cpu.py)

Both scripts can be run directly.

Train the MPR-Net model with default arguments by running

python train.py

Image Deblurring: Evaluation

To evaluate the SSIM for both methods:

• Wiener Filter baseline (./deblur/wiener/eval.py)
• MPRNet deep learning model (./deblur/mprnet/samples/eval.py)

Both scripts can be run directly.

---