Image captioning is one of the primary goals in computer vision which aims to automatically generate natural descriptions for images. Intuitively, human visual system can notice some stimulating regions at first glance, and then volitionally focus on interesting objects within the region. For example, to generate a free-form sentence about “boy-catch-baseball”, the visual region involving “boy” and “baseball” could be first attended and then guide the salient object discovery for the word-by-word generation. Till now, previous captioning works mainly rely on the object-wise modeling and ignore the rich regional patterns. To mitigate the drawback, this paper proposes the region-aware interaction learning method, which aims to explicitly capture the semantic correlations in the region and object dimensions for the word inference. First, given an image, we extract a set of regions which contain diverse objects and their relations. Second, we present the spatial-GCN interaction refining structure which can establish the connection between regions and objects to effectively capture contextual information. Third, we design the dual-attention interaction inference procedure, which enables attention to be calculated in region and object dimensions jointly for the word generation. Specifically, the guidance mechanism is proposed to selectively emphasize semantic inter-dependencies from region to object attentions. Extensive experiments on the MSCOCO dataset demonstrate the superiority of the proposed method. Additional ablation studies and visualization further validate its effectiveness.

External knowledge has been widely applied in image captioning tasks to enrich the generated sentences. However, existing methods retrieve knowledge by considering only semantic relevance while ignoring whether they are useful for captioning. For example, when querying “person” in external knowledge, the most relevant concepts may be “wearing shirt” or “riding horse” statistically, which are not consistent with image contents and introduce noise to generated sentences. Intuitively, we humans can iteratively correlate visual clues with corresponding knowledge to distinguish useful clues from noise. Therefore, we propose an event-aware retrospective learning network for knowledge-based image captioning, which employs a retrospective validation mechanism on captioning models to align the retrieved knowledge with visual contents. This approach is an event-aware perspective and helps select useful knowledge that corresponds to visual facts. To better align images and knowledge, 1) we design an event-aware retrieval algorithm that clusters word-centered knowledge into triplet-centered knowledge (i.e., from “< subject - predicate - object >” to “< triplet A> - edge - < triplet B >”, which provides an event context to facilitate knowledge retrieval and validation. 2) We revisit image contents to retrospectively validate retrieved knowledge by aligning the visual representation between knowledge and image. We summarize the visual characteristics of each knowledge event from the visual genome dataset to help learn which knowledge does not exist in the visual scene and should be discarded. 3) We adopt a dynamic knowledge fusion module that calibrates image and knowledge representations for sentence generation, which includes a knowledge-controlled gate unit that jointly calculates visual and semantic features in event-aware patterns. Compared to current knowledge-based captioning methods, the proposed network retrospectively learns the visual facts by event-aware retrieval and knowledge-image visual alignment, which regularizes the knowledge-incorporated captioning with visual evidence. Extensive experiments on the MS-COCO dataset demonstrate the effectiveness of our method. Ablation studies and visualization demonstrate the advantages of each component of the proposed model.

Multimodal large language models (MLLMs) integrate sophisticated large vision models (LVMs) to empower large language models (LLMs) with vision ability to perceive, reason, and interact in vision-language (V-L) tasks, while the modality bridge between two specialists becomes the bottleneck that translates visual signals into linguistic representations. However, most of the existing methods train the modality bridge with coarse-grained image-text pairs, neglecting the structural mapping between V-L semantics that facilitates modality translation from LVMs to LLMs. To mitigate this, we propose a Constituency-Tree-Induced Multimodal Bridging mechanism (CTIMB) that learns the fine-grained connection from LVMs to LLMs by the structural guidance from multi-modal constituency tree. Our approach consists of: 1) the multi-modal constituency-tree parser that jointly exploits the semantic structure of vision and language; 2) the lightweight connector that translates visual signals into linguistic representation and re-arranges them according to the constituency-tree structure; 3) the dynamic construction loss that aids in aligning the semantic structures derived from the tree parser and the connector. The CTIMB can learn the fine-grained mapping between visual and linguistic semantics, seamlessly bridge the LVMs and LLMs to enhance V-L tasks, and is more cost-efficient compared with current methods. Extensive experiments have demonstrated that our method more accurately interprets the visual features, enabling LLMs to conduct downstream tasks more effectively, and achieve superior performance with less training cost.