List your company
ResearchFormerly WizWand

Image Classification on CUB-200

91.1AccuracyBioVITA
51.767261.978672.1982.4014May 17, 2016Jan 23, 2018Oct 1, 2019Jun 8, 2021Feb 14, 2023Oct 22, 2024Jul 1, 2026
Updated 4mo ago

Evaluation results

MethodLinks
BioVITA2026.0391.1——
CoAt-CBMInterpretability=Yes2026.0489.13——
SAMixBackbone=ResNeXt-502021.1186.83——
CBM-SuiteBottleneck type=CBMs with soft concepts, Backbone=ViT2026.0786.73——
AutoMixBackbone=ResNeXt-502021.1186.56——
ManifoldMixBackbone=ResNeXt-502021.1186.38——
CutMixBackbone=ResNeXt-502021.1185.68——
Linear ProbeInterpretability=No2026.0485.48——
ResizeMixBackbone=ResNeXt-502021.1184.77——
MixupBackbone=ResNeXt-502021.1184.58——
PuzzleMixBackbone=ResNeXt-502021.1184.51——
Concept ConesBackbone=DINO2025.1284.39——
HybridCBMInterpretability=Yes2026.0484.25——
HybridCBMBottleneck type=CBMs with soft concepts, Backbone=ViT2026.0784.25——
ViT-B2024.0384.1——
FMixBackbone=ResNeXt-502021.1184.06——
DOT-CBMInterpretability=Yes2026.0483.76——
DC-ViT-Blatency reduction=24.8%, compression=few-shot2024.0383.5——
SaliencyMixBackbone=ResNeXt-502021.1183.29——
VanillaBackbone=ResNeXt-502021.1183.01——
ResCBMInterpretability=Yes2026.0482.12——
LaboInterpretability=Yes2026.0481.96——
SAMixBackbone=ResNet-182021.1181.11——
XBMBottleneck type=Explanation bottleneck, Protocol=text + image classifier2026.0780.99——
TENETBackbone=ResNet-50, Samples Per Class (SPC)=302021.0380.34——
Sparse-CBM2024.0480.02——
AutoMixBackbone=ResNet-182021.1179.87——
ManifoldMixBackbone=ResNet-182021.1179.76——
Linear Probing2024.0479.29——
Label-FreeInterpretability=Yes2026.0479.01——
PuzzleMixBackbone=ResNet-182021.1178.63——
ResizeMixBackbone=ResNet-182021.1178.5——
CutMixBackbone=ResNet-182021.1178.4——
MixupBackbone=ResNet-182021.1178.39——
SaliencyMixBackbone=ResNet-182021.1177.95——
GLICOBackbone=ResNet-50, Samples Per Class (SPC)=302021.0377.75——
VanillaBackbone=ResNet-182021.1177.68——
FMixBackbone=ResNet-182021.1177.28——
CaBMBottleneck type=Strictly text bottleneck, Protocol=text-only classifier2026.0776.92——
TENETBackbone=ResNet-50, Samples Per Class (SPC)=202021.0376.91——
DeiT-TGFLOPS=1.262023.0576.8——
StandardSparse final layer=false2025.0276.7——
SALF-CBMSparse final layer=false2025.0276.21——
StandardSparse final layer=true2025.0275.96——
Taxabind2026.0375——
CutoutBackbone=ResNet-50, Samples Per Class (SPC)=302021.0374.97——
l1-CBM2024.0474.91——
Zero-TPGFLOPS=0.91, Backbone=DeiT-T2023.0574.4——
SALF-CBMSparse final layer=true2025.0274.35——
LF-CBMSparse final layer=true2025.0274.31——
Label-Free CBM2024.0474.31——
Label-free CBM2025.1274.31——
GLICOBackbone=ResNet-50, Samples Per Class (SPC)=202021.0374.16——
ATSGFLOPS=0.90, Backbone=DeiT-T2023.0573.8——
Random EraseBackbone=ResNet-50, Samples Per Class (SPC)=302021.0373.74——
ToMeGFLOPS=0.90, Backbone=DeiT-T2023.0572.9——
Concept ConesBackbone=RN502025.1272.79——
LPtype=black-box2024.0472.14——
LaBosupervision=full-supervised2024.0471.8——
MixMatchBackbone=ResNet-50, Samples Per Class (SPC)=302021.0370.41——
SimSiam + AugSelfFine-tuning strategy=L2SP, Pre-trained on=ImageNet100, Backbone=ResNet-502021.1170.22——
PS-CBMBottleneck type=CBMs with soft concepts, Backbone=CNN2026.0770.1——
Concept ConesBackbone=VIT2025.1269.7——
CutoutBackbone=ResNet-50, Samples Per Class (SPC)=202021.0368.47——
VLG-CBMBottleneck type=CBMs with soft concepts, Backbone=CNN2026.0768——
SimSiamFine-tuning strategy=L2SP, Pre-trained on=ImageNet100, Backbone=ResNet-502021.1167.59——
Contrastive-CBM2024.0467.04——
LaBoBottleneck type=CBMs with soft concepts, Backbone=CNN2026.0766.8——
Random EraseBackbone=ResNet-50, Samples Per Class (SPC)=202021.0366.14——
TENETBackbone=ResNet-50, Samples Per Class (SPC)=102021.0366.07——
Concept Matrix Search2024.0465.17——
Original CBMconcept_source=labeled concept annotations2024.0465.13——
GLICOBackbone=ResNet-50, Samples Per Class (SPC)=102021.0365.13——
CutoutBackbone=ResNet-50, Samples Per Class (SPC)=102021.0364.33——
Post-hoc CBMbase_model=CLIP2024.0463.92——
Random EraseBackbone=ResNet-50, Samples Per Class (SPC)=102021.0363.72——
A.T.Backbone=ResNet-50, Samples Per Class (SPC)=302021.0363.67——
PCBM-annotationconcept_bank=textual embeddings of annotated attributes2024.0463.63——
DescriptionCLS2024.0463.46——
Zero-Shot2024.0462.63——
Res-CBM-afterstage=after incremental concept discovery2024.0462.43——
SWEETModel Scale=L6H62026.0161.51——
DMAEModel Scale=L6H62026.0160.84——
MixMatchBackbone=ResNet-50, Samples Per Class (SPC)=202021.0360.57——
Res-CBM-beforestage=initial base concept bank2024.0459.87——
P-CBMSparse final layer=true2025.0259.6——
P-CBM2025.1259.6——
PCBMBottleneck type=CBMs with soft concepts, Backbone=CNN2026.0758.8——
LF-CBMBottleneck type=CBMs with soft concepts, Backbone=CNN2026.0758.6——
A.T.Backbone=ResNet-50, Samples Per Class (SPC)=202021.0357.91——
PCBM-hInterpretability=No2026.0457.71——
DS-SJEZero-shot=true2016.0556.8——
WAVEModel Scale=L6H62026.0156.77——
VS2 + PASSBackbone=ViT-B/16, Zero-shot (no retrieval)=true2025.0656.63——
PCBMInterpretability=Yes2026.0456.21——
VS2Backbone=ViT-B/16, Zero-shot (no retrieval)=true2025.0656.14——
CLIPZSBackbone=ViT-B/16, Zero-shot (no retrieval)=true2025.0655.06——
WT-SelectModel Scale=L6H62026.0154.99——
DA-SJEZero-shot=true2016.0554.3——
SAEF+γ RECBackbone=ViT-B/16, Zero-shot (no retrieval)=true2025.0653.28——