HyperAI초신경

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

한국어

HyperAI초신경

Image Captioning On Nocaps Val Overall

평가 지표

CIDEr

Pretrain (#images)

SPICE

평가 결과

이 벤치마크에서 각 모델의 성능 결과

모델 이름	CIDEr	Pretrain (#images)	SPICE	Paper Title	Repository
BLIP_CapFilt-L	109.6	129M	14.7	BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation
BLIP_ViT-L	113.2	129M	14.8	BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation
BLIP-2 ViT-G FlanT5 XL (zero-shot)	121.6	1.1B	15.8	BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models
Enc-Dec	90.2	-	12.1	Conceptual 12M: Pushing Web-Scale Image-Text Pre-Training To Recognize Long-Tail Visual Concepts
BLIP-2 ViT-G OPT 6.7B (zero-shot)	121.0	1.1B	15.3	BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models
OSCAR	80.9	345M	11.3	Oscar: Object-Semantics Aligned Pre-training for Vision-Language Tasks
VinVL	95.5	5.7M	13.5	VinVL: Revisiting Visual Representations in Vision-Language Models
BLIP-2 ViT-G OPT 2.7B (zero-shot)	119.7	1.1B	15.4	BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models
SimVLM	112.2	1.8B	-	SimVLM: Simple Visual Language Model Pretraining with Weak Supervision
OmniVL	107.5	14M	14.7	OmniVL:One Foundation Model for Image-Language and Video-Language Tasks	-
LEMON_large	113.4	200M	15.0	Scaling Up Vision-Language Pre-training for Image Captioning	-

0 of 11 row(s) selected.