List your company
ResearchFormerly WizWand

Semantic Segmentation on ADE20K 83 (val)

54.7mIoUVAN-B6
32.02837.91443.849.686Feb 20, 2022Feb 27, 2022Mar 7, 2022Mar 15, 2022Mar 22, 2022Mar 30, 2022Apr 7, 2022
Updated 8mo ago

Evaluation results

MethodLinks
VAN-B6Params (M)=231, Input size=512x512, GFLOPs=1658, Pre-trained=ImageNet-22K2022.0254.7
VAN-B5Params (M)=117, Input size=512x512, GFLOPs=1208, Pre-trained=ImageNet-22K2022.0253.9
ConvNeXt-LParams (M)=235, Input size=640x640, GFLOPs=2458, Pre-trained=ImageNet-22K2022.0253.7
Swin-LParams (M)=234, Input size=640x640, GFLOPs=2468, Pre-trained=ImageNet-22K2022.0253.5
ConvNeXt-BParams (M)=122, Input size=640x640, GFLOPs=1828, Pre-trained=ImageNet-22K2022.0253.1
UperNetBackbone=VAN-B4, Parameters (M)=90, GFLOPs=1098, Training Scheme=Following [15], Input Resolution=2048x5122022.0252.2
Swin-BParams (M)=121, Input size=640x640, GFLOPs=1841, Pre-trained=ImageNet-22K2022.0251.7
UperNetBackbone=VAN-B3, Parameters (M)=75, GFLOPs=1030, Training Scheme=Following [15], Input Resolution=2048x5122022.0250.6
UperNetBackbone=VAN-B2, Parameters (M)=57, GFLOPs=948, Training Scheme=Following [15], Input Resolution=2048x5122022.0250.1
UperNetBackbone=ConvNeXt-B, Parameters (M)=122, GFLOPs=1170, Training Scheme=Following [15], Input Resolution=2048x5122022.0249.9
UperNetBackbone=Swin-B, Parameters (M)=121, GFLOPs=1188, Training Scheme=Following [15], Input Resolution=2048x5122022.0249.7
UperNetBackbone=ConvNeXt-S, Parameters (M)=82, GFLOPs=1027, Training Scheme=Following [15], Input Resolution=2048x5122022.0249.5
DaViT-BaseBackbone=DaViT-Base, Segmentation Method=UperNet, #Params (M)=121, FLOPs (G)=1175, Evaluation Strategy=Single-scale2022.0449.4
UperNetBackbone=Swin-S, Parameters (M)=81, GFLOPs=1038, Training Scheme=Following [15], Input Resolution=2048x5122022.0249.3
ViT-HybridBackbone=ViT-Hybrid, Segmentation Method=DPT, #Params (M)=124, FLOPs (G)=1231, Evaluation Strategy=Single-scale2022.0449
Focal-BaseBackbone=Focal-Base, Segmentation Method=UperNet, #Params (M)=126, FLOPs (G)=1354, Evaluation Strategy=Single-scale2022.0449
DaViT-SmallBackbone=DaViT-Small, Segmentation Method=UperNet, #Params (M)=81, FLOPs (G)=1030, Evaluation Strategy=Single-scale2022.0448.8
Twins-SVT-LargeBackbone=Twins-SVT-Large, Segmentation Method=UperNet, #Params (M)=133, FLOPs (G)=1188, Evaluation Strategy=Single-scale2022.0448.8
ResNeSt-200Backbone=ResNeSt-200, Segmentation Method=DLab.v3+, #Params (M)=88, FLOPs (G)=1381, Evaluation Strategy=Single-scale2022.0448.4
XCiT-M24/8Backbone=XCiT-M24/8, Segmentation Method=UperNet, #Params (M)=109, Evaluation Strategy=Single-scale2022.0448.4
Swin-BaseBackbone=Swin-Base, Segmentation Method=UperNet, #Params (M)=121, FLOPs (G)=1188, Evaluation Strategy=Single-scale2022.0448.1
Semantic FPNBackbone=VAN-B3, Parameters (M)=49, GFLOPs=68, Training Scheme=Following [98], Input Resolution=512x5122022.0248.1
Focal-SmallBackbone=Focal-Small, Segmentation Method=UperNet, #Params (M)=85, FLOPs (G)=1130, Evaluation Strategy=Single-scale2022.0448
Twins-SVT-BaseBackbone=Twins-SVT-Base, Segmentation Method=UperNet, #Params (M)=88, FLOPs (G)=1044, Evaluation Strategy=Single-scale2022.0447.7
Swin-SmallBackbone=Swin-Small, Segmentation Method=UperNet, #Params (M)=81, FLOPs (G)=1038, Evaluation Strategy=Single-scale2022.0447.6
Semantic FPNBackbone=PVTv2-B3, Parameters (M)=49, GFLOPs=62, Training Scheme=Following [98], Input Resolution=512x5122022.0247.3
XCiT-S24/16Backbone=XCiT-S24/16, Segmentation Method=UperNet, #Params (M)=73, Evaluation Strategy=Single-scale2022.0446.9
HamNetBackbone=ResNet-101, Parameters (M)=69, GFLOPs=1111, Training Scheme=Following [15], Input Resolution=2048x5122022.0246.8
Semantic FPNBackbone=VAN-B2, Parameters (M)=30, GFLOPs=48, Training Scheme=Following [98], Input Resolution=512x5122022.0246.7
UperNetBackbone=ConvNeXt-T, Parameters (M)=60, GFLOPs=939, Training Scheme=Following [15], Input Resolution=2048x5122022.0246.7
DaViT-TinyBackbone=DaViT-Tiny, Segmentation Method=UperNet, #Params (M)=60, FLOPs (G)=940, Evaluation Strategy=Single-scale2022.0446.3
Twins-SVT-SmallBackbone=Twins-SVT-Small, Segmentation Method=UperNet, #Params (M)=54, FLOPs (G)=912, Evaluation Strategy=Single-scale2022.0446.2
UperNetBackbone=Swin-T, Parameters (M)=60, GFLOPs=945, Training Scheme=Following [15], Input Resolution=2048x5122022.0246.1
XCiT-S12/16Backbone=XCiT-S12/16, Segmentation Method=UperNet, #Params (M)=52, Evaluation Strategy=Single-scale2022.0445.9
Focal-TinyBackbone=Focal-Tiny, Segmentation Method=UperNet, #Params (M)=62, FLOPs (G)=998, Evaluation Strategy=Single-scale2022.0445.8
HRNet-w48Backbone=HRNet-w48, Segmentation Method=OCRNet, #Params (M)=71, FLOPs (G)=664, Evaluation Strategy=Single-scale2022.0445.7
OCRNetBackbone=ResNet-101, Parameters (M)=56, GFLOPs=923, Training Scheme=Following [15], Input Resolution=2048x5122022.0245.3
Semantic FPNBackbone=PVTv2-B2, Parameters (M)=29, GFLOPs=46, Training Scheme=Following [98], Input Resolution=512x5122022.0245.2
ResNet-101Backbone=ResNet-101, Segmentation Method=UperNet, #Params (M)=86, FLOPs (G)=1029, Evaluation Strategy=Single-scale2022.0444.9
UperNetBackbone=ResNet-101, Parameters (M)=86, GFLOPs=1029, Training Scheme=Following [15], Input Resolution=2048x5122022.0244.9
PVT-LargeBackbone=PVT-Large, Segmentation Method=SemanticFPN, #Params (M)=65, FLOPs (G)=318, Evaluation Strategy=Single-scale2022.0444.8
Swin-TinyBackbone=Swin-Tiny, Segmentation Method=UperNet, #Params (M)=60, FLOPs (G)=945, Evaluation Strategy=Single-scale2022.0444.5
Semantic FPNBackbone=PVT-Medium, Parameters (M)=48, GFLOPs=61, Training Scheme=Following [98], Input Resolution=512x5122022.0243.5
Semantic FPNBackbone=VAN-B1, Parameters (M)=18, GFLOPs=35, Training Scheme=Following [98], Input Resolution=512x5122022.0242.9
Semantic FPNBackbone=PVTv2-B1, Parameters (M)=18, GFLOPs=34, Training Scheme=Following [98], Input Resolution=512x5122022.0242.5
Semantic FPNBackbone=PoolFormer-S36, Parameters (M)=35, GFLOPs=48, Training Scheme=Following [98], Input Resolution=512x5122022.0242
Semantic FPNBackbone=PoolFormer-S24, Parameters (M)=23, GFLOPs=39, Training Scheme=Following [98], Input Resolution=512x5122022.0240.3
Semantic FPNBackbone=PVT-Small, Parameters (M)=28, GFLOPs=45, Training Scheme=Following [98], Input Resolution=512x5122022.0239.8
Semantic FPNBackbone=ResNeXt101-32x4d, Parameters (M)=47, GFLOPs=65, Training Scheme=Following [98], Input Resolution=512x5122022.0239.7
Semantic FPNBackbone=ResNet101, Parameters (M)=48, GFLOPs=65, Training Scheme=Following [98], Input Resolution=512x5122022.0238.8
Semantic FPNBackbone=VAN-B0, Parameters (M)=8, GFLOPs=26, Training Scheme=Following [98], Input Resolution=512x5122022.0238.5
Semantic FPNBackbone=PVTv2-B0, Parameters (M)=8, GFLOPs=25, Training Scheme=Following [98], Input Resolution=512x5122022.0237.2
Semantic FPNBackbone=PoolFormer-S12, Parameters (M)=16, GFLOPs=31, Training Scheme=Following [98], Input Resolution=512x5122022.0237.2
Semantic FPNBackbone=ResNet50, Parameters (M)=29, GFLOPs=46, Training Scheme=Following [98], Input Resolution=512x5122022.0236.7
Semantic FPNBackbone=PVT-Tiny, Parameters (M)=17, GFLOPs=33, Training Scheme=Following [98], Input Resolution=512x5122022.0235.7
Semantic FPNBackbone=ResNet18, Parameters (M)=16, GFLOPs=32, Training Scheme=Following [98], Input Resolution=512x5122022.0232.9