Datasets:
- https://github.com/BathVisArtData/PhotoArt50
- https://github.com/BathVisArtData/PeopleArt
- http://www.robots.ox.ac.uk/~vgg/data/paintings/
Reading:
- YOLOv2 https://pjreddie.com/darknet/yolo/
- YOLO https://arxiv.org/pdf/1506.02640.pdf
- Faster R-CNN https://arxiv.org/pdf/1506.01497.pdf
- RCNN history https://blog.athelas.com/a-brief-history-of-cnns-in-image-segmentation-from-r-cnn-to-mask-r-cnn-34ea83205de4
- Neural Style Transfer https://medium.com/artists-and-machine-intelligence/neural-artistic-style-transfer-a-comprehensive-look-f54d8649c199
- Neural Style Transfer Review https://arxiv.org/pdf/1705.04058.pdf
Project Idea: Detecting objects in artwork with R-CNN, Fast R-CNN, and YOLO/YOLOv2 (in PyTorch?)
- Start with pre-trained model and report results on PeopleArt, PhotoArt50
- Try training on PhotoArt/PeopleArt.
- Try "removing" style with neural style transfer from painting to photograph.
- Try "removing" style by tuning parameters in neural style transfer.
Presentation Goal:
- Have pre-trained results of some R-CNN variant and YOLO on PeopleArt and PhotoArt50
- (Hopefully) train one of R-CNN variant and YOLO on PeopleArt or PhotoArt50
- Have basic neural style removal
Current progress:
- Have some neural style removal, not good in general, but have one figure and can explain what we've tried and what we can keep trying
- Have DataLoader for PeopleArt, PeopleArt photos
- (David) Working on YOLO pre-trained - Pytorch version is really finicky... maybe I'll try the tensorflow version...