Inference Performance and Resource Usage on Semantic Segmentation on Cityscapes
7.3FPSSegNeXt-L
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SegNeXt-LParams=48.8M, FLOPs=554G2026.03 | 7.3 | 16.9 | |
| SegFormer-B3Params=47.3M, FLOPs=963G2026.03 | 6.6 | 21.9 | |
| ARTA-TinyParams=49.5M, FLOPs=286±20G2026.03 | 6.6 | 14.6 | |
| ARTA-SmallParams=61.7M, FLOPs=355±25G2026.03 | 6 | 19.8 | |
| SegMAN-BParams=51.8M, FLOPs=479G2026.03 | 5.6 | 23.8 | |
| ARTA-BaseParams=111.5M, FLOPs=590±40G2026.03 | 4.9 | 25.4 | |
| SegFormer-B4Params=64.1M, FLOPs=1241G2026.03 | 4.7 | 29.5 | |
| SegFormer-B5Params=84.7M, FLOPs=1460G2026.03 | 4 | 34.9 | |
| SegMAN-LParams=92.4M, FLOPs=796G2026.03 | 3.9 | 31.6 | |
| AFF-TinyParams=46.5M, FLOPs=463G2026.03 | 3.8 | 34.4 | |
| AFF-SmallParams=62.1M, FLOPs=639G2026.03 | 3.6 | 38.7 |