List your company
ResearchFormerly WizWand

Image Classification on ImageNet-1k 1.0 (val) (Accuracy)

0.848Top-1 AccuracyAdventurer-Base/P8
0.716960.750980.7850.81902Oct 10, 2024
Updated 8mo ago

Evaluation results

MethodLinks
Adventurer-Base/P8Token mixer=Mamba, Input size=448, Parameters=100M, Throughput (↑)=57, Memory (↓)=>80G2024.100.848
MambaReg-LToken mixer=Mamba, Input size=384, Parameters=342M, Throughput (↑)=23, Memory (↓)=>80G2024.100.845
Adventurer-BaseToken mixer=Mamba, Input size=448, Parameters=99M, Throughput (↑)=216, Memory (↓)=45.2G2024.100.843
MambaReg-BToken mixer=Mamba, Input size=384, Parameters=99M, Throughput (↑)=63, Memory (↓)=51.4G2024.100.843
EfficientNet-B7Token mixer=2D Conv., Input size=560, Parameters=66M, Throughput (↑)=61, Memory (↓)=>80G2024.100.843
DeiT-Base/P14 (our impl.)Token mixer=Self-Attention, Input size=560, Parameters=87M, Throughput (↑)=41, Memory (↓)=>80G2024.100.84
VMamba-BToken mixer=Mamba + 2D Conv., Input size=224, Parameters=89M, Throughput (↑)=246, Memory (↓)=37.1G2024.100.839
ConvNeXt-BToken mixer=2D Conv., Input size=224, Parameters=89M, Throughput (↑)=305, Memory (↓)=17.9G2024.100.838
MambaReg-LToken mixer=Mamba, Input size=224, Parameters=341M, Throughput (↑)=67, Memory (↓)=55.5G2024.100.836
DeiT-Base/P14 (our impl.)Token mixer=Self-Attention, Input size=448, Parameters=86M, Throughput (↑)=86, Memory (↓)=>80G2024.100.835
Adventurer-LargeToken mixer=Mamba, Input size=224, Parameters=346M, Throughput (↑)=301, Memory (↓)=35.5G2024.100.834
ConvNeXt-SToken mixer=2D Conv., Input size=224, Parameters=50M, Throughput (↑)=412, Memory (↓)=13.1G2024.100.831
DeiT-BaseToken mixer=Self-Attention, Input size=384, Parameters=86M, Throughput (↑)=201, Memory (↓)=63.8G2024.100.831
MambaReg-BToken mixer=Mamba, Input size=224, Parameters=99M, Throughput (↑)=196, Memory (↓)=20.3G2024.100.83
Adventurer-BaseToken mixer=Mamba, Input size=224, Parameters=99M, Throughput (↑)=856, Memory (↓)=13.0G2024.100.826
ConvNeXt-TToken mixer=2D Conv., Input size=224, Parameters=29M, Throughput (↑)=635, Memory (↓)=8.3G2024.100.821
Adventurer-SmallToken mixer=Mamba, Input size=224, Parameters=44M, Throughput (↑)=1405, Memory (↓)=8.3G2024.100.818
DeiT-BaseToken mixer=Self-Attention, Input size=224, Parameters=86M, Throughput (↑)=861, Memory (↓)=14.4G2024.100.818
EfficientNet-B3Token mixer=2D Conv., Input size=300, Parameters=12M, Throughput (↑)=546, Memory (↓)=19.7G2024.100.816
MambaReg-SToken mixer=Mamba, Input size=224, Parameters=28M, Throughput (↑)=391, Memory (↓)=9.9G2024.100.814
Vim-SmallToken mixer=Mamba, Input size=224, Parameters=26M, Throughput (↑)=395, Memory (↓)=9.4G2024.100.805
DeiT-SmallToken mixer=Self-Attention, Input size=224, Parameters=22M, Throughput (↑)=1924, Memory (↓)=6.8G2024.100.798
Adventurer-TinyToken mixer=Mamba, Input size=224, Parameters=12M, Throughput (↑)=2757, Memory (↓)=4.2G2024.100.782
MambaReg-TToken mixer=Mamba, Input size=224, Parameters=9M, Throughput (↑)=746, Memory (↓)=5.1G2024.100.774
Vim-TinyToken mixer=Mamba, Input size=224, Parameters=7M, Throughput (↑)=750, Memory (↓)=4.8G2024.100.761
DeiT-TinyToken mixer=Self-Attention, Input size=224, Parameters=5M, Throughput (↑)=3480, Memory (↓)=3.4G2024.100.722