I would like to know if there exists papers combining SIFT algorithm with Deep Learning networks. I'm thinking of something similar to extract points with SIFT and get a descriptor of those point's neighborhood with a CNN. Thanks in advance.
The following methods may benefit you:
1- Use some corner detection methods (like Harris, Susan, etc.) to find locations of points of interest. After that you can open a window around each point to extract the sub-image which can finally be an input to CNN.
2. Use SIFT descriptors as input to CNN. As you know that CNN works only with 2-D representation (like images), so you must firstly re-represent SIFT descriptors as a 2-D data (i.e. convert the descriptors into image with some width and height which depends on your conversion method).
Key.Net: Keypoint Detection by Handcrafted and Learned CNN Filters and
GLAMpoints: Greedily Learned Accurate Match points are two papers I have come across.
GLAMpoints does the other way around extract keypoints through CNN and use SIFT descriptors
- Badges
- Science topic
- Similar topics
- Mathematics
- Applied Mathematics