Osprey: Pixel Understanding with Visual Instruction Tuning


Osprey: Pixel Understanding with Visual Instruction Tuning

Abstract masked dataset 추가로 제안 ⇒ pixel-wise understanding 위해. design a vision-language model by injecting pixel-level representation into LLM CNN based CLIP for using as a image encoder and mask-aware visual extractor Applications; They use also with SAM for more semantic works. Introduction previous research limitation. Region-level understanding에 국한되어 있는 연구들 언급. Kosmos-2 [37]: 이 연구는 bounding box를 지정된 영역으로 처리하고 객체 수준의 공간적 특징을 활용하는 시각적 지시 조정을 시도했습니다. [ https://github.com/microsoft/unilm/tree/mas...


#AI #SOTA #Similarity #SentencesBERT #SAM #RegionLevel #PixelWise #Pixelunderstanding #Osprey #mLLM #InstructionFollowing #HQSAM #GPT4Gen #generative #explainable #Deeplearning #Dataset #BERT #VisualInstructionTuning

원문링크 : Osprey: Pixel Understanding with Visual Instruction Tuning