태그 목록
- 전체보기(70)
- Paper(34)
- study(28)
- Graph(23)
- CS224W(17)
- 3D(14)
- Point Review(12)
- 3D Geometry(10)
- Object Detection(8)
- Scene graph(7)
- Generative Model(6)
- segmentation(4)
- tracking(4)
- 360(2)
- Siam(2)
- transformer(2)
- 중요 개념(1)
- graphics(1)
- Plane(1)
- Calibration(1)
- Knowledge(1)
- Depth Estimation(1)
- Reconstruction(1)
- View Synthesis(1)
- Tangent image(1)
- Room layout(1)
- NLP(1)
- Layout Estimation(1)
- NeRF(1)
- Panorama(1)
- Deformable(1)
- PBR(1)
- Recommendation(1)
- Representation(1)
- Classification(1)
전체보기 (70)Paper(34)study(28)Graph(23)CS224W(17)3D(14)Point Review(12)3D Geometry(10)Object Detection(8)Scene graph(7)Generative Model(6)segmentation(4)tracking(4)360(2)Siam(2)transformer(2)중요 개념(1)graphics(1)Plane(1)Calibration(1)Knowledge(1)Depth Estimation(1)Reconstruction(1)View Synthesis(1)Tangent image(1)Room layout(1)NLP(1)Layout Estimation(1)NeRF(1)Panorama(1)Deformable(1)PBR(1)Recommendation(1)Representation(1)Classification(1)
[논문 리뷰] Understanding Contrastive Representation Learning through Alignment and Uniformity on the Hypersphere
Representation을 학습하는데 있어 $l2$ norm을 통해 feature의 영역을 unit hypershpere으로 제한하는 것은 학습 안정성을 높이고 클래스를 적절히 분류하여 선형 분리가 가능하도록 만든다. $l2$ norm은 보편적인 방법이지만 encod
2022년 1월 16일
·
0개의 댓글