*STA9132
-
Essay 6 - REPA*STA9132/Essays 2025. 11. 24. 20:56
마지막 essay!!으헤헤 신난다. Essay 졸업! 굳이 공들여 쓸 필요가 없었을 수도 있다. 애초에 교수님께서 엄청난 걸 요구하신 것도 아니고. 정말 간단한 essay 정도를 말씀하신 것 같은데..(근데 summury, criticism 이런 기본 윤곽을 정해주셨는데, 아무래도 1/2 ~ 1page는 뻥... 이셨던.. 것만 같은..) 내 욕심에, 공부한 거 정리하고, 관련된 거 찾아보고 하다보니 다소 burden은 되었지만.그래도 나름 굵직한 주제들 다시 복습하고, 정리도 되고. 보람있었다.
-
[REPA] Representation Alignment for Generation: Training DiT*STA9132/Essays 2025. 11. 23. 19:42
https://sihyun.me/REPA/ Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You ThinkRepresentation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Thinksihyun.me
-
[I-JEPA] Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture*STA9132/Essays 2025. 11. 23. 15:54
https://ai.meta.com/blog/yann-lecun-ai-model-i-jepa/ The first AI model based on Yann LeCun’s vision for more human-like AIIllustrating how the predictor learns to model the semantics of the world. For each image, the portion outside of the blue box is encoded and given to the predictor as context. The predictor outputs a representation for what it expects to be in the regionai.meta.comhttps://a..
-
-
[DINOv2] Learning Robust Visual Features without Supervision*STA9132/Essays 2025. 11. 23. 08:48
https://ai.meta.com/blog/dino-v2-computer-vision-self-supervised-learning/ DINOv2: State-of-the-art computer vision models with self-supervised learningDINOv2 is able to take a video and generate a higher-quality segmentation than the original DINO method. DINOv2 allows remarkable properties to emerge, such as a robust understanding of object parts, and robust semantic and low-level understandin..
-
[SiT] Exploring Flow and Diffusion-basedGenerative Models with Scalable InterpolantTransformers*STA9132/Essays 2025. 11. 22. 22:40
https://scalable-interpolant.github.io/ https://scalable-interpolant.github.io/We present Scalable interpolant Transformers (SiT), a family of generative models built on the backbone of Diffusion Transformers (DiT). The interpolant framework, which allows for connecting two distributions in a more flexible way than standard diffusionscalable-interpolant.github.io