Novel seam carving method
My method steps:
1. Pre-scaling
2. Create energy map
a- Create gradient map
I used Bi-Directional Cascade Network for Perceptual Edge Detection, CVPR 2019 and this pytorch implementation.
Download the pre-trained model here.
b- Create saliency map
I used Pyramid Feature Attention Network for Saliency Detection, CVPR 2019 and this pytorch implementation.
Download the pre-trained model here.
c- Use depth map
I assumed that the depth map is given as input to the system.
3. Create energy map: GM + SM + DM
4. Select the appropriate threshold for the transition from seam carving to scaling
5. Seam carving
6. Post-scaling
Results
Original image | 50 % width reduction | 60 % | 70 % | 80 % | 90 % |
---|---|---|---|---|---|
Support
Contact me @:
e-mail:
Telegram id: