chunked/nltk_chunking/model_doc_detr/chunk_25.txt

These input embeddings are learnt positional encodings that the authors refer to as object queries, and similarly to
the encoder, they are added to the input of each attention layer.