List your company
ResearchFormerly WizWand

Image Captioning on TextCaps

164.3CIDErPaLI-3
59.15686.453113.75141.047Oct 7, 2022May 17, 2023Dec 25, 2023Aug 4, 2024Mar 14, 2025Oct 22, 2025Jun 2, 2026
Updated 4mo ago

Evaluation results

MethodLinks
PaLI-3OCR pipeline input=true, Resolution=812x8122023.10164.3
PaLI-XOCR pipeline input=true, Resolution=812x8122023.10163.7
PaLIInput type=Pipelined2022.10160.4
PaLI-3OCR pipeline input=false, Resolution=812x8122023.10158.8
PaLI-XOCR pipeline input=false, Resolution=812x8122023.10147
GIT2Input type=Pixel-only, Additional labeled data=true2022.10145
GenLIPArch=g/16, Data=8.0B2026.05135.4
SigLIP2Arch=g/16, Data=40.0B2026.05134.5
DocOwl2Size=8B, Visual tokens=3242024.09131.8
DocOwl 1.5Size=8B, Visual tokens=16982024.09131.6
SigLIP2Arch=So/16, Data=40.0B2026.05131.5
GenLIPArch=L/16, Data=8.0B2026.05131.4
GenLIPArch=So/16, Data=8.0B2026.05129.5
LLaVA-1.5 + QA-ViTObserved in training=true2024.02128.7
SigLIP2Arch=L/16, Data=40.0B2026.05127.8
OVision2Arch=L/16, Data=12.8B2026.05127.4
LLaVA-1.5Observed in training=true2024.02126.4
InstructBLIPObserved in training=true, OCR tokens=true2024.02126
LLaVA-NeXT+CALLLM=Vicuna-7B, Resolution setting=High resolution setting2024.05124.7
LLaVA-NeXT+CALLLM=Vicuna-13B, Resolution setting=High resolution setting2024.05124.4
AIMv2Arch=L/14, Data=12.0B2026.05122.4
SigLIPArch=L/16, Data=40.0B2026.05120.7
UReaderSize=7B, Visual tokens=8412024.09118.4
LLaVA-NeXTLLM=Vicuna-13B, Resolution setting=High resolution setting, Win/All=4/42024.05118.2
CLIPArch=L/14, Data=12.8B2026.05117.9
LLaVA-NeXTLLM=Vicuna-7B, Resolution setting=High resolution setting, Win/All=4/42024.05115
mPLUG-DocOwlObserved in training=true2024.02111.9
DocOwlSize=7B, Visual tokens=8412024.09111.9
LION-12Bparameters=12B2023.11108.76
LLaVA-1.5+CALLLM=Vicuna-7B, Resolution setting=Low resolution setting2024.05106.7
LBRBackbone=LLaVA-NEXT-3B2026.05105.8
LLaVA-1.5+CALLLM=Vicuna-13B, Resolution setting=Low resolution setting2024.05105.6
InstructBLIP (T5XXL)backbone=T5XXL2023.11105.44
LBRBackbone=LLaVA-1.5-13B2026.05105.2
LION-4Bparameters=4B2023.11104.87
InstructBLIP (T5XL)backbone=T5XL2023.11104.17
LLaVA-NEXT-3BBackbone=LLaVA-NEXT-3B2026.05104
LLaVA-1.5-13BBackbone=LLaVA-1.5-13B2026.05103.9
LLaVA-1.5LLM=Vicuna-13B, Resolution setting=Low resolution setting, Win/All=3/42024.05102.4
LLaVA-1.5LLM=Vicuna-7B, Resolution setting=Low resolution setting, Win/All=4/42024.05100.7
VanillaModel=LLaVA-1.5-7B, Retention Ratio=50%2026.06100.3
VanillaModel=LLaVA-1.5-7B, Retention Ratio=20%2026.06100.3
VanillaModel=LLaVA-1.5-7B, Retention Ratio=10%2026.06100.3
TGV-KVModel=LLaVA-1.5-7B, Retention Ratio=50%2026.0699.8
TAPInput type=Pipelined2022.1099.5
LBRBackbone=LLaVA-1.5-7B2026.0599.1
LLaVA-1.5-7BBackbone=LLaVA-1.5-7B2026.0598.4
MM1-30BShot=16, Model Scale=30B2024.0397.6
MM1-30BShot=162024.0397.6
SnapKVModel=LLaVA-1.5-7B, Retention Ratio=50%2026.0697.1
Pix2Struct-LargeInput type=Pixel-only, Scale=Large2022.1095.5
Pix2Struct-largeSize=1B, Visual tokens=2048, Fine-tuned=true2024.0995.5
PrefixKVModel=LLaVA-1.5-7B, Retention Ratio=50%2026.0694.3
MonkeyParameters=9.8B, Resolution=896x8962023.1193.2
MonkeySize=9B, Visual tokens=12802024.0993.2
MM1-7BShot=162024.0393.1
MM1-30BShot=8, Model Scale=30B2024.0392.9
MM1-30BShot=82024.0392.9
MM1-3BShot=162024.0391.6
MM1-3BShot=8, Model Scale=3B2024.0388.8
MM1-3BShot=82024.0388.8
MM1-7BShot=8, Model Scale=7B2024.0388.2
MM1-7BShot=82024.0388.2
Pix2Struct-BaseInput type=Pixel-only, Scale=Base2022.1088
Pix2Struct-baseSize=<1B, Visual tokens=2048, Fine-tuned=true2024.0988
TGV-KVModel=LLaVA-1.5-7B, Retention Ratio=20%2026.0687.8
MM1-30BShot=42024.0387.5
MM1-7BShot=42024.0384.5
MM1-3BShot=42024.0384.1
AIMv2architecture=ViT-3B/14, #patches=576, resolution=336px2024.1183.8
AIMv2architecture=ViT-1B/14, #patches=576, resolution=336px2024.1182.6
InstructBLIP+ (T5XL)backbone=T5XL, uses_in-house_data=true2023.1182.55
InstructBLIP+ (T5XXL)backbone=T5XXL, uses_in-house_data=true2023.1182.53
SnapKVModel=LLaVA-1.5-7B, Retention Ratio=20%2026.0682.3
IDEFICS-80BShot=16, Model Scale=80B2024.0381.4
IDEFICS-80BShot=162024.0381.4
AIMv2architecture=ViT-L/14, #patches=576, resolution=336px2024.1181.1
M4CInput type=Pipelined2022.1081
AIMv2architecture=ViT-H/14, #patches=576, resolution=336px2024.1180.7
MiniGPTV2evaluated_by_authors=true2023.1180.6
StreamingLLMModel=LLaVA-1.5-7B, Retention Ratio=50%2026.0678.4
IDEFICS-80BShot=8, Model Scale=80B2024.0377.6
IDEFICS-80BShot=82024.0377.6
InstructBLIPObserved in training=true2024.0275.6
OpenAI CLIParchitecture=ViT-L/14, #patches=5762024.1175.3
DonutInput type=Pixel-only2022.1074.4
IDEFICS-80BShot=42024.0372.7
SigLIParchitecture=ViT-So400M/14, #patches=7522024.1170.8
BLIP22024.0270.4
SigLIParchitecture=ViT-L/14, #patches=5762024.1169.9
IDEFICS-9BShot=162024.0367.4
Qwen-VLBackbone=Qwen-7B2023.1165.1
MM1-30BShot=0, Model Scale=30B2024.0364.9
MM1-30BShot=02024.0364.9
MM1-7BShot=0, Model Scale=7B2024.0364.2
MM1-7BShot=02024.0364.2
TGV-KVModel=LLaVA-1.5-7B, Retention Ratio=10%2026.0663.6
MM1-3BShot=0, Model Scale=3B2024.0363.3
MM1-3BShot=02024.0363.3
IDEFICS-9BShot=8, Model Scale=9B2024.0363.2