Semantic Segmentation on ADE20K 80k iterations (val)
44.6mIoU (%)FastViT-MA36
Evaluation Results
| Method | Links | |
|---|---|---|
| FastViT-MA36Backbone=FastViT-MA36, GPU Latency (ms)=8.2, Mobile Latency (ms)=16.3, Parameters (M)=45.7, FLOPs (G)=53, Decoder=Semantic FPN, Crop Size=512x512, Batch Size=22023.03 | 44.6 | |
| FastViT-SA36Backbone=FastViT-SA36, GPU Latency (ms)=6.1, Mobile Latency (ms)=12.9, Parameters (M)=33.6, FLOPs (G)=44, Decoder=Semantic FPN, Crop Size=512x512, Batch Size=22023.03 | 42.9 | |
| PoolFormer-M36Backbone=PoolFormer-M36, GPU Latency (ms)=41.4, Mobile Latency (ms)=24.8, Parameters (M)=59.8, FLOPs (G)=68, Decoder=Semantic FPN, Crop Size=512x512, Batch Size=22023.03 | 42.4 | |
| PoolFormer-S36Backbone=PoolFormer-S36, GPU Latency (ms)=28.0, Mobile Latency (ms)=17.0, Parameters (M)=34.6, FLOPs (G)=48, Decoder=Semantic FPN, Crop Size=512x512, Batch Size=22023.03 | 42 | |
| PVT-MediumBackbone=PVT-Medium, Parameters (M)=48.0, FLOPs (G)=55, Decoder=Semantic FPN, Crop Size=512x512, Batch Size=22023.03 | 41.6 | |
| FastViT-SA24Backbone=FastViT-SA24, GPU Latency (ms)=4.4, Mobile Latency (ms)=9.3, Parameters (M)=23.8, FLOPs (G)=37, Decoder=Semantic FPN, Crop Size=512x512, Batch Size=22023.03 | 41 | |
| PoolFormer-S24Backbone=PoolFormer-S24, GPU Latency (ms)=18.9, Mobile Latency (ms)=10.7, Parameters (M)=23.2, FLOPs (G)=48, Decoder=Semantic FPN, Crop Size=512x512, Batch Size=22023.03 | 40.3 | |
| PVT-SmallBackbone=PVT-Small, Parameters (M)=28.2, FLOPs (G)=41, Decoder=Semantic FPN, Crop Size=512x512, Batch Size=22023.03 | 39.8 | |
| ResNet-101Backbone=ResNet-101, GPU Latency (ms)=4.6, Mobile Latency (ms)=12.4, Parameters (M)=47.5, FLOPs (G)=65, Decoder=Semantic FPN, Crop Size=512x512, Batch Size=22023.03 | 38.8 | |
| FastViT-SA12Backbone=FastViT-SA12, GPU Latency (ms)=2.5, Mobile Latency (ms)=5.6, Parameters (M)=14.1, FLOPs (G)=29, Decoder=Semantic FPN, Crop Size=512x512, Batch Size=22023.03 | 38 | |
| PoolFormer-S12Backbone=PoolFormer-S12, GPU Latency (ms)=9.7, Mobile Latency (ms)=6.2, Parameters (M)=15.7, FLOPs (G)=31, Decoder=Semantic FPN, Crop Size=512x512, Batch Size=22023.03 | 37.2 | |
| ResNet-50Backbone=ResNet-50, GPU Latency (ms)=2.7, Mobile Latency (ms)=8.7, Parameters (M)=28.5, FLOPs (G)=46, Decoder=Semantic FPN, Crop Size=512x512, Batch Size=22023.03 | 36.7 |