A night-to-day image translation implementation using Pix2Pix. The model was trained on the University of Maryland's DNIM dataset for 100 epochs, and actually achieved a pretty good result (at least visually). Credit goes to affinelayer for the tensorflow pix2pix package. Check out Day2Night.ipynb for the code!
Evaluating Local Features for Day-Night Matching - University of Maryland, College Park, USA.
Git clonethis repository (through Github Desktop or through the link).- Open Day2Night.ipynb
- Run codeblocks sequentially until
!python pix2pix.py \.
- Go to this Colab link
- Click Copy To Drive
- Run codeblocks sequentially until
!python pix2pix.py \.
Follow the instructions on the notebook.
- Scroll down & run
!python pix2pix.py \ --mode train \ - OPTIONAL: Change input/output folders and training Epochs.
# test the model
!python pix2pix.py \
--mode test \
--output_dir daynight_test \
--input_dir daynight/val \
--checkpoint daynight_train# show images in notebook
import matplotlib.pyplot as plt
import numpy as np
choice = random.randint(0,150)
path = "/content/pix2pix-tensorflow/daynight_test/images/"
fig = plt.figure(figsize=(14,18))
axarr = fig.subplots(1,3)
[ax.axis('off') for ax in axarr]
axarr[0].imshow(Image.open(path + ('image%d'%choice) + '-inputs.png'))
axarr[0].set_title("Input",size=15,color='white',weight='bold')
axarr[1].imshow(Image.open(path + ('image%d'%choice) + '-outputs.png'))
axarr[1].set_title("Reconstruction",size=15,color='white',weight='bold')
axarr[2].imshow(Image.open(path + ('image%d'%choice) + '-targets.png'))
axarr[2].set_title("Original",size=15,color='white',weight='bold')
plt.show()
.png){kind=link}
.png){kind=link}
.png){kind=link}
