HyperAI초신경

홈 뉴스 최신 연구 논문 튜토리얼 데이터셋 백과사전 SOTA LLM 모델 GPU 랭킹 컨퍼런스

한국어

시스템 설정

HyperAI초신경

Image To Text Retrieval On Coco

평가 지표

Recall@1

Recall@10

Recall@5

평가 결과

이 벤치마크에서 각 모델의 성능 결과

모델 이름	Recall@1	Recall@10	Recall@5	Paper Title	Repository
BLIP-2 (ViT-L, fine-tuned)	83.5	98.0	96.0	BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models
BLIP-2 (ViT-G, fine-tuned)	85.4	98.5	97.0	BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models
DVSA	-	74.8	-	Deep Visual-Semantic Alignments for Generating Image Descriptions
ONE-PEACE (ViT-G, w/o ranking)	84.1	98.3	96.3	ONE-PEACE: Exploring One General Representation Model Toward Unlimited Modalities
Unicoder-VL	-	97.2	-	Unicoder-VL: A Universal Encoder for Vision and Language by Cross-modal Pre-training	-
SigLIP (ViT-L, zero-shot)	70.6	-	-	Sigmoid Loss for Language Image Pre-Training
Oscar	-	99.8	-	Oscar: Object-Semantics Aligned Pre-training for Vision-Language Tasks
FLAVA (ViT-B, zero-shot)	42.74	-	76.76	FLAVA: A Foundational Language And Vision Alignment Model
IAIS	67.78	94.48	89.7	Learning Relation Alignment for Calibrated Cross-modal Retrieval
CLIP (zero-shot)	58.4	88.1	81.5	Learning Transferable Visual Models From Natural Language Supervision

0 of 10 row(s) selected.