LXMERT¶
Overview¶
The LXMERT model was proposed in LXMERT: Learning Cross-Modality Encoder Representations from Transformers by Hao Tan & Mohit Bansal. It is a series of bidirectional transformer encoders (one for the vision modality, one for the language modality, and then one to fuse both modalities) pretrained using a combination of masked language modeling, visual-language text alignment, ROI-feature regression, masked visual-attribute modeling, masked visual-object modeling, and visual-question answering objectives. The pretraining consists of multiple multi-modal datasets: MSCOCO, Visual-Genome + Visual-Genome Question Answering, VQA 2.0, and GQA.
The abstract from the paper is the following:
Vision-and-language reasoning requires an understanding of visual concepts, language semantics, and, most importantly, the alignment and relationships between these two modalities. We thus propose the LXMERT (Learning Cross-Modality Encoder Representations from Transformers) framework to learn these vision-and-language connections. In LXMERT, we build a large-scale Transformer model that consists of three encoders: an object relationship encoder, a language encoder, and a cross-modality encoder. Next, to endow our model with the capability of connecting vision and language semantics, we pre-train the model with large amounts of image-and-sentence pairs, via five diverse representative pretraining tasks: masked language modeling, masked object prediction (feature regression and label classification), cross-modality matching, and image question answering. These tasks help in learning both intra-modality and cross-modality relationships. After fine-tuning from our pretrained parameters, our model achieves the state-of-the-art results on two visual question answering datasets (i.e., VQA and GQA). We also show the generalizability of our pretrained cross-modality model by adapting it to a challenging visual-reasoning task, NLVR, and improve the previous best result by 22% absolute (54% to 76%). Lastly, we demonstrate detailed ablation studies to prove that both our novel model components and pretraining strategies significantly contribute to our strong results; and also present several attention visualizations for the different encoders
Tips:
Bounding boxes are not necessary to be used in the visual feature embeddings, any kind of visual-spacial features will work.
Both the language hidden states and the visual hidden states that LXMERT outputs are passed through the cross-modality layer, so they contain information from both modalities. To access a modality that only attends to itself, select the vision/language hidden states from the first input in the tuple.
The bidirectional cross-modality encoder attention only returns attention values when the language modality is used as the input and the vision modality is used as the context vector. Further, while the cross-modality encoder contains self-attention for each respective modality and cross-attention, only the cross attention is returned and both self attention outputs are disregarded.
The original code can be found here.