CPTR: Full Transformer Network for Image Captioning
This work addresses image captioning for computer vision researchers by introducing a convolution-free approach, though it is incremental as it builds on existing Transformer architectures.
The paper tackles image captioning by proposing CPTR, a full Transformer network that processes sequentialized raw images, eliminating the need for CNNs. The model outperforms conventional CNN+Transformer methods on the MSCOCO dataset.
In this paper, we consider the image captioning task from a new sequence-to-sequence prediction perspective and propose CaPtion TransformeR (CPTR) which takes the sequentialized raw images as the input to Transformer. Compared to the "CNN+Transformer" design paradigm, our model can model global context at every encoder layer from the beginning and is totally convolution-free. Extensive experiments demonstrate the effectiveness of the proposed model and we surpass the conventional "CNN+Transformer" methods on the MSCOCO dataset. Besides, we provide detailed visualizations of the self-attention between patches in the encoder and the "words-to-patches" attention in the decoder thanks to the full Transformer architecture.