Ideas

[eric-yyjau]

For class project

How to use filter banks to approximate neural networks?

References

Learning a Discriminative Filter Bank Within a CNN for Fine-Grained Recognition http://openaccess.thecvf.com/content_cvpr_2018/papers/Wang_Learning_a_Discriminative_CVPR_2018_paper.pdf

Action Recognition With Spatial-Temporal Discriminative Filter Banks http://openaccess.thecvf.com/content_ICCV_2019/papers/Martinez_Action_Recognition_With_Spatial-Temporal_Discriminative_Filter_Banks_ICCV_2019_paper.pdf

[eric-yyjau]

Feature extractor

[PDF] R2D2: Repeatable and Reliable Detector and Descriptor - Semantic Scholar https://www.semanticscholar.org/paper/R2D2%3A-Repeatable-and-Reliable-Detector-and-Revaud-Weinzaepfel/911aa843612a8a47579e4d3da141fdee564965ba

Key.Net: Keypoint Detection by Handcrafted and Learned CNN Filters axelBarroso/Key.Net: Code for the ICCV19 paper Key.Net: Keypoint Detection by Handcrafted and Learned CNN Filters https://github.com/axelBarroso/Key.Net

DFOB: Detecting and describing features by octagon filter bank for fast image matching - ScienceDirect https://www.sciencedirect.com/science/article/pii/S0923596515002155

previous references

Wavelet Feature Selection for Image Classification - IEEE Journals & Magazine https://ieeexplore.ieee.org/document/4586427

[eric-yyjau]

Proposal

Problem definition Image denoising using u-net, using filter banks.

Add gaussian noise to the images.

Current approach

Proposed approach

References

[yiqian-wang]

Bilateral filter bank for denoising http://www.ece.lsu.edu/ipl/papers/IEEE_TIP2008.pdf

[eric-yyjau]

Proposal 1:

Noisy images break the sift. Filter banks filter out the images and feed into sift. It works.

Proposal 2:

superpoint is data hungry. Can we add filter banks to denoise and make superpoint works better with less training data?

[eric-yyjau]

Gym

• cycling + O-bike unlock
• where are the O-bike?