This document discusses generative adversarial networks (GANs) and their relationship to reinforcement learning. It begins with an introduction to GANs, explaining how they can generate images without explicitly defining a probability distribution by using an adversarial training process. The second half discusses how GANs are related to actor-critic models and inverse reinforcement learning in reinforcement learning. It explains how GANs can be viewed as training a generator to fool a discriminator, similar to how policies are trained in reinforcement learning.
This document summarizes recent research on applying self-attention mechanisms from Transformers to domains other than language, such as computer vision. It discusses models that use self-attention for images, including ViT, DeiT, and T2T, which apply Transformers to divided image patches. It also covers more general attention modules like the Perceiver that aims to be domain-agnostic. Finally, it discusses work on transferring pretrained language Transformers to other modalities through frozen weights, showing they can function as universal computation engines.
【DLゼミ】XFeat: Accelerated Features for Lightweight Image Matching
公開URL:https://arxiv.org/pdf/2404.19174
出典:Guilherme Potje, Felipe Cadar, Andre Araujo, Renato Martins, Erickson R. ascimento: XFeat: Accelerated Features for Lightweight Image Matching, Proceedings of the 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2023)
概要:リソース効率に優れた特徴点マッチングのための軽量なアーキテクチャ「XFeat(Accelerated Features)」を提案します。手法は、局所的な特徴点の検出、抽出、マッチングのための畳み込みニューラルネットワークの基本的な設計を再検討します。特に、リソースが限られたデバイス向けに迅速かつ堅牢なアルゴリズムが必要とされるため、解像度を可能な限り高く保ちながら、ネットワークのチャネル数を制限します。さらに、スパース下でのマッチングを選択できる設計となっており、ナビゲーションやARなどのアプリケーションに適しています。XFeatは、高速かつ同等以上の精度を実現し、一般的なラップトップのCPU上でリアルタイムで動作します。
This document discusses generative adversarial networks (GANs) and their relationship to reinforcement learning. It begins with an introduction to GANs, explaining how they can generate images without explicitly defining a probability distribution by using an adversarial training process. The second half discusses how GANs are related to actor-critic models and inverse reinforcement learning in reinforcement learning. It explains how GANs can be viewed as training a generator to fool a discriminator, similar to how policies are trained in reinforcement learning.
This document summarizes recent research on applying self-attention mechanisms from Transformers to domains other than language, such as computer vision. It discusses models that use self-attention for images, including ViT, DeiT, and T2T, which apply Transformers to divided image patches. It also covers more general attention modules like the Perceiver that aims to be domain-agnostic. Finally, it discusses work on transferring pretrained language Transformers to other modalities through frozen weights, showing they can function as universal computation engines.
【DLゼミ】XFeat: Accelerated Features for Lightweight Image Matchingharmonylab
公開URL:https://arxiv.org/pdf/2404.19174
出典:Guilherme Potje, Felipe Cadar, Andre Araujo, Renato Martins, Erickson R. ascimento: XFeat: Accelerated Features for Lightweight Image Matching, Proceedings of the 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2023)
概要:リソース効率に優れた特徴点マッチングのための軽量なアーキテクチャ「XFeat(Accelerated Features)」を提案します。手法は、局所的な特徴点の検出、抽出、マッチングのための畳み込みニューラルネットワークの基本的な設計を再検討します。特に、リソースが限られたデバイス向けに迅速かつ堅牢なアルゴリズムが必要とされるため、解像度を可能な限り高く保ちながら、ネットワークのチャネル数を制限します。さらに、スパース下でのマッチングを選択できる設計となっており、ナビゲーションやARなどのアプリケーションに適しています。XFeatは、高速かつ同等以上の精度を実現し、一般的なラップトップのCPU上でリアルタイムで動作します。
4. まとめた論文の一覧 (1/3)
~PointNet
1. STN: M. Jaderberg, et al.. Spatial Transformer Networks. NIPS2015. (2015-01-05)
2. PointNet: C. R. Qi, et al.. PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation. CVPR2017.
(2016-12-02)
入力にボクセルを利用
3. OctNet: G. Riegler, et al.. OctNet: Learning Deep 3D Representations at High Resolutions. CVPR2017. (2016-11-15)
4. O-CNN: P. –S. Wang, et al.. O-CNN: Octree-based Convolutional Neural Networks for 3D Shape Analysis. SIGGRAPH2017.
(2017-07-03)
5. RSNet: Q. Huang, et al.. Recurrent Slice Networks for 3D Segmentation of Point Clouds. CVPR2018. (2018-02-13)
6. PointGrid: T. Le, et al.. PointGrid: A Deep Network for 3D Shape Understanding. CVPR2018. (2018-06-18)
7. AO-CNN: P. –S. Wang, et al.. Adaptive O-CNN: A Patch-based Deep Representation of 3D Shapes. SIGGRAPH Asia2018.
(2018-09-19)
入力を点群・ボクセル以外に変換
8. FPNN: Y. Li, et al.. FPNN: Field Probing Neural Networks for 3D Data. NIPS2016. (2016-05-20)
9. Kd-Network: R. Klokov, et al.. Escape from Cells: Deep Kd-Networks for The Recognition of 3D Point Cloud Models.
ICCV2017. (2017-04-04)
10. PPFNet: H. Deng, et al.. PPFNet: Global Context Aware Local Features for Robust 3D Point Matching. CVPR2018.
(2018-02-07)
11. SO-Net: J. Li, et al.. SO-Net: Self-Organizing Network for Point Cloud Analysis. CVPR2018. (2018-03-12)
12. MCCNN: P. Hermosilla, et al.. Monte Carlo Convolution for Learning on Non-Uniformly Sampled Point Clouds.
SIGGRAPH Asia2018. (2018-06-05)
13. Automatic Depth Image Generation: R. Roveri, et al.. A Network Architecture for Point Cloud Classification via Automatic
Depth Images Generation. CVPR2018. (2018-06-18)
14. MRTNet: M. Gadelha, et al.. Multiresolution Tree Networks for 3D Point Cloud Processing. ECCV2018. (2018-07-10)
Google Spreadsheetsで公開しています
https://docs.google.com/spreadsheets/d/1MJTdLHIRefclx4GkZkfZxFj9GIxLSjit_f32j_DR9gw/edit?usp=sharing
5. まとめた論文の一覧 (2/3)
局所形状の利用
15. ECC: M. Simonovsky, et al.. Dynamic Edge-Conditioned Filters in Convolutional Neural Networks on Graphs. CVPR2017.
(2017-04-10)
16. PointNet++: C. R. Qi, et al.. PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space. NIPS2017.
(2017-06-07)
17. Pointwise CNN: B. –S. Hua, et al.. Pointwise Convolutional Neural Networks. CVPR2018. (2017-12-14)
18. KCN: Y. Shen, et al.. Mining Point Cloud Local Structures by Kernel Correlation and Graph Pooling. CVPR2018. (2017-12-19)
19. PointCNN: Y. Li, et al.. PointCNN. arXiv:1801.07791. (2018-01-23)
20. DG-CNN: Y. Wang, et al.. Dynamic Graph CNN for Learning on Point Clouds. arXiv:1801.07829. (2018-01-24)
21. Flex-Convolution: F. Groh, et al.. Flex-Convolution. (Million-Scale Point-Cloud Learning Beyond Grid-Worlds). ACCV2018.
(2018-03-20)
22. PCNN: M. Atzmon, et al.. Point Convolutional Neural Networks by Extension Operators. SIGGRAPH2018. (2018-03-27)
23. SpiderCNN: Y. Xu, et al.. SpiderCNN: Deep Learning on Point Sets with Parameterized Convolutional Filters. ECCV2018.
(2018-03-30)
24. Parametric Continuous Convolutions: S. Wang, et al.. Deep Parametric Continuous Convolutional Neural Networks.
CVPR2018. (2018-06-18)
25. Tangent Convolution: M. Tatarchenko, et al.. Tangent Convolutions for Dense Prediction in 3D.
CVPR2018. (2018-06-18)
26. SCN: S. Xie, et al.. Attentional ShapeContextNet for Point Cloud Recognition. CVPR2018. (2018-06-18)
Google Spreadsheetsで公開しています
https://docs.google.com/spreadsheets/d/1MJTdLHIRefclx4GkZkfZxFj9GIxLSjit_f32j_DR9gw/edit?usp=sharing
6. まとめた論文の一覧 (3/3)
特殊なレイヤー
27. SEGCloud: L. P. Tchapmi, et al.. SEGCloud: Semantic Segmentation of 3D Point Clouds. 3DV2017. (2017-10-20)
28. FoldingNet: Y. Yang, et al.. FoldingNet: Point Cloud Auto-encoder via Deep Grid Deformation. CVPR2018. (2017-12-19)
29. SPLATNet: H. Su, et al.. SPLATNet: Sparse Lattice Networks for Point Cloud Processing. CVPR2018. (2018-02-22)
30. URSA: M. B. Skouson. URSA: A Neural Network for Unordered Point Clouds Using Constellations. arXiv:1808.04848.
(2018-08-14)
31. Point Cloud VAE-GAN: C. Kingkan, et al.. Generating Mesh-based Shapes From Learned Latent Spaces of Point Clouds
with VAE-GAN. ICPR2018. (2018-08-20)
32. Fully Convolutional Point Network: D. Rethage, et al.. Fully-Convolutional Point Networks for Large-Scale Point Clouds.
ECCV2018. (2018-08-21)
33. PPF-FoldingNet: H. Deng, et al.. PPF-FoldNet: Unsupervised Learning of Rotation Invariant 3D Local Descriptors. ECCV2018.
(2018-08-30)
個別のアプリケーション
34. PCPNet: P. Guerrero, et al.. PCPNET: Learning Local Shape Properties from Raw Point Clouds. Eurographics2018. (2017-10-13)
35. Frustum PointNet: C. R. Qi, et al.. Frustum PointNets for 3D Object Detection from RGB-D Data. CVPR2018. (2017-11-22)
36. SGPN: W. Wang, et al.. SGPN: Similarity Group Proposal Network for 3D Point Cloud Instance Segmentation. CVPR2018.
(2017-11-23)
37. PU-Net: L. Yu, et al.. PU-Net: Point Cloud Upsampling Network. CVPR2018. (2018-01-21)
38. Hand PointNet: L. Ge, et al.. Hand PointNet: 3D Hand Pose Estimation Using Point Sets. CVPR2018. (2018-06-18)
39. Point Attention Network: C. Kingkan, et al.. Point Attention Network for Gesture Recognition Using Point Cloud Data.
BMVC2018. (2018-09-03)
40. P2P Reg PointNet: L. Ge, et al.. Point-to-Point Regression PointNet for 3D Hand Pose Estimation. ECCV2018. (2018-10-08)
まだたくさんあるとは思いますが,今回は以上の論文を扱います
分類も難しかったのですが,とりあえず以上のように分けました
Google Spreadsheetsで公開しています
https://docs.google.com/spreadsheets/d/1MJTdLHIRefclx4GkZkfZxFj9GIxLSjit_f32j_DR9gw/edit?usp=sharing