Week 2 Meeting Notes

This stuff just goes over what we discussed in the second week of meetings for sketch2drawing.

Notes From Last Meeting

Steps

• Get Data
• Augment/Modify Data
• Edge Detection On Data to Generate Sketch (??)
• Train GAN
• Do some cross validation and tune hyperparameters
• Test GAN

Possible

• Landscape
• Food
• Cartoon (difficult to do edge detection on, possibly find sketch version)

Kaggle Datasets

https://www.kaggle.com/arnaud58/landscape-pictures

http://natural-scenes.cps.utexas.edu/db.shtml

https://www.kaggle.com/kmader/food41

https://www.kaggle.com/ktaebum/anime-sketch-colorization-pair

Possible Architectures

• CycleGAN
• StarGAN
• U-nets
• Conditional GANs
• DCGANs

Steps

• Tune thresholds for Canny
• Data Collection
• Pre-processing