Table 5 Ablation study on the two improvement modules.
From: An efficient and accurate 2D human pose estimation method using VTTransPose network
Model | Backbone | Params (Mb) | Memory (batch size = 4) (Mb) | AP (coco val gt bbox) |
|---|---|---|---|---|
TransPose-H–S | HRNet-S-W32 | 8 | 3503 | 76.1 |
TransPose-H–S + twin attention | HRNet-S-W32 | 8 | 1953 | 76.3 |
VTTransPose | HRNet-S-W32 + V block | 6 | 2007 | 76.5 |
