PointTransformer: Encoding Human Local Features for Small Target Detection

Figure 3

For each local feature, both self-attention and cross-attention mechanisms are used to reconstruct the local features of the human body.