List your company
ResearchFormerly WizWand

Document Image Classification on RVL-CDIP 1.0 (test)

96.17AccuracyDocFormer_BASE
89.035690.887892.7494.5922Jan 29, 2018Oct 12, 2018Jun 26, 2019Mar 9, 2020Nov 20, 2020Aug 4, 2021Apr 18, 2022
Updated 8mo ago

Evaluation results

MethodLinks
DocFormer_BASEParameters=183M, Modality=T+L+I (G), Image Embedding=ResNet-502022.0496.17
StructuralLM_LARGEParameters=355M, Modality=T+L, Image Embedding=None2022.0496.08
LayoutLMv3_LARGEParameters=368M, Modality=T+L+I (P), Image Embedding=Linear2022.0495.93
LiLT_BASEModality=T+L, Image Embedding=ResNeXt101-FPN2022.0495.68
LayoutLMv2_LARGEParameters=426M, Modality=T+L+I (G), Image Embedding=ResNeXt101-FPN2022.0495.64
TILT_LARGEParameters=780M, Modality=T+L+I (R), Image Embedding=U-Net2022.0495.52
DocFormer_LARGEParameters=536M, Modality=T+L+I (G), Image Embedding=ResNet-502022.0495.5
LayoutLMv3_BASEParameters=133M, Modality=T+L+I (P), Image Embedding=Linear2022.0495.44
TILT_BASEParameters=230M, Modality=T+L+I (R), Image Embedding=U-Net2022.0495.25
LayoutLMv2_BASEParameters=200M, Modality=T+L+I (G), Image Embedding=ResNeXt101-FPN2022.0495.25
UDocParameters=272M, Modality=T+L+I (R), Image Embedding=ResNet-502022.0495.05
LayoutLM_BASEParameters=160M, Modality=T+L+I (R), Image Embedding=ResNet-101 (fine-tune)2022.0494.42
SelfDocModality=T+L+I (R), Image Embedding=ResNeXt-1012022.0492.81
Stacked GeneralizationMeta-classifier=MLNN, Model Components=Stacking of holistic & region-based models, Transfer Learning=Inter and intra-domain weights transfer2018.0192.21
LayoutLM_LARGEParameters=343M, Modality=T+L, Image Embedding=None2022.0491.9
Intra-Domain Transfer LearningBackbone=VGG-16, Input=Full document images, Transfer Learning=Weights transfer from VGG-16 trained on ImageNet2018.0191.11
Afzal et al.Ensemble Backbones=AlexNet, VGG-16, GoogLeNet and ResNet-50, Transfer Learning=Weights transfer2018.0190.97
Tensmeyer et al.Backbone=AlexNet, Architecture=Spatial Pyramidal Pooling (SPP), Resolution=384 x 384, Aspect Ratio Preservation=true2018.0190.94
Csurka et al.Backbone=GoogLeNet, Transfer Learning=ImageNet based transfer learning2018.0190.7
RoBERTa_LARGEParameters=355M, Modality=T, Image Embedding=None2022.0490.11
RoBERTa_BASEParameters=125M, Modality=T, Image Embedding=None2022.0490.06
BERT_LARGEParameters=340M, Modality=T, Image Embedding=None2022.0489.92
BERT_BASEParameters=110M, Modality=T, Image Embedding=None2022.0489.81
Harley et al.Backbone=AlexNet, Approach=Document section-based models, Transfer Learning=Weight transfer from Alexnet, Ensemble=max voting2018.0189.8
Tensmeyer et al.Backbone=AlexNet, Architecture=Spatial Pyramidal Pooling (SPP), Transfer Learning=None2018.0189.31