Notice
Recent Posts
Recent Comments
Link
| 일 | 월 | 화 | 수 | 목 | 금 | 토 |
|---|---|---|---|---|---|---|
| 1 | 2 | 3 | ||||
| 4 | 5 | 6 | 7 | 8 | 9 | 10 |
| 11 | 12 | 13 | 14 | 15 | 16 | 17 |
| 18 | 19 | 20 | 21 | 22 | 23 | 24 |
| 25 | 26 | 27 | 28 | 29 | 30 | 31 |
Tags
- PnP algorithm
- prompt learning
- Energy-based model
- Continual Learning
- Class Incremental
- Mask diffusion
- L2P
- Mask-and-replace diffusion strategy
- DualPrompt
- CVPR2022
- Vector Quantized Diffusion Model for Text-to-Image Synthesis
- timm
- Discrete diffusion
- VQ-VAE
- state_dict()
- Face Pose Estimation
- mmcv
- ENERGY-BASED MODELS FOR CONTINUAL LEARNING
- learning to prompt for continual learning
- Facial Landmark Localization
- Class Incremental Learning
- CIL
- learning to prompt
- Markov transition matrix
- requires_grad
- Face Alignment
- img2pose: Face Alignment and Detection via 6DoF
- Img2pose
- VQ-diffusion
- 베이지안 정리
Archives
- Today
- Total
Computer Vision , AI
[One-page summary] DualPrompt: Complementary Prompting for Rehearsal-free Continual Learning (ECCV 2022) By Wang et al. 본문
Paper_review[short]
[One-page summary] DualPrompt: Complementary Prompting for Rehearsal-free Continual Learning (ECCV 2022) By Wang et al.
Elune001 2023. 4. 18. 17:42●Summary: Learning two disjoint prompt spaces makes rehearsal-free prompt-based continual learning more effectively
●Approach highlight
○Using task-invariant General prompt(L_g) at ViT 1~2nd layer and task-specific Expert Prompt(# of task ×L_e) at ViT 3~5th layer
○Prefix tuning: before MSA, concatenate key prompt to hidden representation of key and value prompt to hidden representation of value, respectively
●Main Results:
●Discussion
○Can be generalized? (Isn’t it specialized for the ViT?)
'Paper_review[short]' 카테고리의 다른 글
'Paper_review[short]' Related Articles
more
