Study With Inha

Meta FAIR SAM 2: Segment Anything in Images and Videos Segment Anything Model 2 설명 및 논문 리뷰Paper: SAM2 Paper LinkDemo: https://sam2.metademolab.comCode: https://github.com/facebookresearch/segment-anything-2 Website: https://ai.meta.com/sam2 Meta Segment Anything Model 2Access our research Open innovation To enable the research community to build upon this work, we’re publicly releasing a pretrai..

CVPR 2024 Highlight Paper(FG-OVD) The devil is in the fine-grained details: Evaluating open-vocabulary object detectors for fine-grained understanding논문 링크: https://arxiv.org/abs/2311.17518프로젝트 페이지: https://lorebianchi98.github.io/FG-OVD/ The devil is in the fine-grained details: Evaluating open-vocabulary object detectors for fine-grained understandingWe evaluated some state-of-the-art models o..

ICLR 2023 Spotlight (notable-top-25%),(SparK) Designing BERT for Convolutional Networks: Sparse and Hierarchical Masked Modeling논문 링크: https://openreview.net/forum?id=NRxydtWup1SGitHub: https://github.com/keyu-tian/SparK Designing BERT for Convolutional Networks: Sparse and Hierarchical...This paper presents a simple yet powerful framework to pre-train convolutional network (convnet) with Sparse..

ICLR 2024 Oral Paper,Is ImageNet worth 1 video?Learning strong image encoders from 1 long unlabelled video논문 링크: https://openreview.net/forum?id=Yen1lGns2o Is ImageNet worth 1 video? Learning strong image encoders from 1...Self-supervised learning has unlocked the potential of scaling up pretraining to billions of images, since annotation is unnecessary. But are we making the best use of data? H..

CVPR 2024 Highlight Paper, Adapting Visual-Language Models for Generalizable Anomaly Detection in Medical Images 논문 리뷰Paper Link: https://arxiv.org/pdf/2403.12570GitHub Link: https://github.com/MediaBrain-SJTU/MVFA-AD 1. Introduction여러가지 Modality를 하나의 모델로 다루는 연구가 지속됨에 따라서, 최근 Visual-Language Model (VLM)이 괄목할만한 성능을 보이고 있다.하지만 CLIP을 포함한 대부분의 VLM은 대용량의 Natural image들로 pretrain되었기에, Domain Gap이나 Tas..

CVPR 2024, Open-World Semantic Segmentation Including Class Similarity논문 링크: https://www.ipb.uni-bonn.de/wp-content/papercite-data/pdf/sodano2024cvpr.pdfGitHub 링크: https://github.com/PRBonn/ContMAV 1. Introduction컴퓨터 비전 기술이 실제 산업에 많이 적용됨에 따라서 조금더 넓은 coverage를 가진 기술에 대한 관심이 많아지고 있다.본 논문도 컴퓨터 비전 task의 coverage를 넓히는 것을 목표로 하는 논문으로, Open World에서 발생하는 문제를 해결하고자 한다.Closed world: 정해져 있는 class 중에 정답을 맞추..

ICLR 2024, Interpreting CLIP's Image Representation via Text-Based Decomposition 논문 링크: https://openreview.net/attachment?id=5Ca9sSzuDp&name=pdf프로젝트 페이지 링크: https://yossigandelsman.github.io/clip_decomposition/ 1. Introduction최근 많은 논문들에서 거대한 Text, Image Pair로 학습시킨 CLIP 모델을 활용하는 후속 연구들을 진행하고 있다.본 논문에서는 CLIP의 이미지 인코더를 분석하여 모델의 각 구성 요소가 final representation에 미치는 영향을 해석 가능한 텍스트로 설명하고 있다.먼저, Attentio..

TMLR 2024, LLM-grounded Diffusion : Enhancing Prompt Understanding of Text-to-Image Diffusion Models with Large Language Models 논문 링크: https://arxiv.org/pdf/2305.13655.pdf 프로젝트 페이지 링크: https://llm-grounded-diffusion.github.io/ 1. Introduction Diffusion Model과 같은 Text-to-Image 생성 모델들의 발전은 높은 자유도로 질 좋은 이미지를 얻을 수 있도록 했다. 하지만 Stable Diffusion 및 최근에 나온 SDXL조차도 Text의 세부적인 내용을 파악하여 정확한 이미지를 생성하는 데에 한계가..

Google Research, NeurIPS 2023 acceptedStableRep: Synthetic Images from Text-to-Image ModelsMake Strong Visual Representation Learners논문 링크: https://arxiv.org/pdf/2306.00984.pdf StableRep은 NeurIPS 2023에 accept된 논문으로 LG AI Research에서 정리한 NeurIPS 2023 주요 연구주제에 선정된 논문이다.LG AI 리서치 블로그: https://www.lgresearch.ai/blog/view?seq=379 [NeurIPS 2023] 주요 연구 주제와 주목할 만한 논문 소개 - LG AI Research BLOGNeurIPS 2023,..

Segment Anything in High Quality, ETH Zurich 논문링크: https://arxiv.org/abs/2306.01567 Segment Anything in High QualityThe recent Segment Anything Model (SAM) represents a big leap in scaling up segmentation models, allowing for powerful zero-shot capabilities and flexible prompting. Despite being trained with 1.1 billion masks, SAM's mask prediction quality falls short inarxiv.org Introduction올해 상..

CVPR 2023 accepted paper, TCL (Text-grounded Contrastive Learning): Learning to Generate Text-grounded Mask for Open-world Semantic Segmentation from Only Image-Text Pairs 논문링크: https://arxiv.org/abs/2212.00785 Learning to Generate Text-grounded Mask for Open-world Semantic Segmentation from Only Image-Text Pairs We tackle open-world semantic segmentation, which aims at learning to segment arbit..

BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation 논문 링크: https://arxiv.org/abs/2201.12086 BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation Vision-Language Pre-training (VLP) has advanced the performance for many vision-language tasks. However, most existing pre-trained models only excel..