object queries input problem

[yangfan293]

Hello, you mentioned in your paper that "object queries into Q and the depth embeddings into K and V by linear layers" in your cross-attention, But I see in your open source code that object queries are entered directly into self.cross_attn_depth after nn.embeding. Neither object queries nor the depth embeddings go through a linear process layers, is there something wrong with me? Thank you for your time.