chunked/nltk_chunking/_tasks_explained/chunk_94.txt

There are two components to the object detection head: a linear layer to transform the decoder hidden states into logits over the class labels, and a MLP to predict the bounding box.