CITE: Compact Interactive TransformEr for Multilingual Image Captioning

VD-SAN: Visual-Densely Semantic Attention Network for Image Caption Generation

CNN-LSTM based architectures have played an important role in image captioning, but limited by the training efficiency and expression ability, researchers began to explore the CNN-Transformer based models and achieved great success. Meanwhile, almost all recent works adopt Faster R-CNN as the backbone encoder to extract region-level features from given images. However, Faster R-CNN needs a pre-training on an additional dataset, which divides the image captioning task into two stages and limits its potential applications. In this paper, we build a pure Transformer-based model, which integrates image captioning into one stage and realizes end-to-end training. Firstly, we adopt SwinTransformer to replace Faster R-CNN as the backbone encoder to extract grid-level features from given images; Then, referring to Transformer, we build a refining encoder and a decoder. The refining encoder refines the grid features by capturing the intra-relationship between them, and the decoder decodes the refined features into captions word by word. Furthermore, in order to increase the interaction between multi-modal (vision and language) features to enhance the modeling capability, we calculate the mean pooling of grid features as the global feature, then introduce it into refining encoder to refine with grid features together, and add a pre-fusion process of refined global feature and generated words in decoder. To validate the effectiveness of our proposed model, we conduct experiments on MSCOCO dataset. The experimental results compared to existing published works demonstrate that our model achieves new state-of-the-art performances of 138.2% (single model) and 141.0% (ensemble of 4 models) CIDEr scores on 'Karpathy' offline test split and 136.0% (c5) and 138.3% (c40) CIDEr scores on the official online test server. Trained models and source code will be released.

Image captioning refers to automatic generation of descriptive texts according to the visual content of images. It is a technique integrating multiple disciplines including the computer vision (CV), natural language processing (NLP) and artificial intelligence. In recent years, substantial research efforts have been devoted to generate image caption with impressive progress. To summarize the recent advances in image captioning, we present a comprehensive review on image captioning, covering both traditional methods and recent deep learning-based techniques. Specifically, we first briefly review the early traditional works based on the retrieval and template. Then deep learning-based image captioning researches are focused, which is categorized into the encoder-decoder framework, attention mechanism and training strategies on the basis of model structures and training manners for a detailed introduction. After that, we summarize the publicly available datasets, evaluation metrics and those proposed for specific requirements, and then compare the state of the art methods on the MS COCO dataset. Finally, we provide some discussions on open challenges and future research directions.

Explaining an image with missing or non-existent objects is known as object bias (hallucination) in image captioning. This behaviour is quite common in the state-of-the-art captioning models which is not desirable by humans. To decrease the object hallucination in captioning, we propose three simple yet efficient training augmentation method for sentences which requires no new training data or increase in the model size. By extensive analysis, we show that the proposed methods can significantly diminish our models’ object bias on hallucination metrics. Moreover, we experimentally demonstrate that our methods decrease the dependency on the visual features. All of our code, configuration files and model weights are available online¹.

Image captioning achieves its goal by two steps traditionally, it first extracts image features using deep networks, and then translates the features into correct text descriptions. Though recent compelling captioning results have surged continuously, the bridge between visual cues and texts is usually limited by weak single-layer CNN features and one-word one-attention strategy. In this paper, we propose a simple yet effective Hierarchical Encoder-And-Decoder image CAPtioning model (HeadCap) to capture visual cues across different levels whilst conducting one-word multi-attention scheme in a unified hierarchical architecture. In specific, we collect multi-layer features from feature pyramid network (FPN), which allows stringing visual features across different levels hierarchically. More importantly, a multi-attention LSTM module (mLSTM) is established on such hierarchical features, which performs multiple attentions over hierarchical features progressively when predicting a word. This provides LSTM more attention chance to reduce the risk of the mismatch between visual cues and texts, and further to bridge the gap between each other. Experiments on MSCOCO image captioning dataset show that HeadCap yields favorable performance against several well-behaved deep captioning models.

Self-attention mechanism, which has been successfully applied to current encoder-decoder framework of image captioning, is used to enhance the feature representation in the image encoder and capture the most relevant information for the language decoder. However, most existing methods will assign attention weights to all candidate vectors, which implicitly hypothesizes that all vectors are relevant. Moreover, current self-attention mechanisms ignore the intra-object attention distribution, and only consider the inter-object relationships. In this paper, we propose a Multi-Gate Attention (MGA) block, which expands the traditional self-attention by equipping with additional Attention Weight Gate (AWG) module and Self-Gated (SG) module. The former constrains the attention weights to be assigned to the most contributive objects. The latter is adopted to consider the intra-object attention distribution and eliminate the irrelevant information in object feature vector. Furthermore, most current image captioning methods apply the original transformer designed for natural language processing task, to refine image features directly. Therefore, we propose a pre-layernorm transformer to simplify the transformer architecture and make it more efficient for image feature enhancement. By integrating MGA block with pre-layernorm transformer architecture into the image encoder and AWG module into the language decoder, we present a novel Multi-Gate Attention Network (MGAN). The experiments on MS COCO dataset indicate that the MGAN outperforms most of the state-of-the-art, and further experiments on other methods combined with MGA blocks demonstrate the generalizability of our proposal.

Unsupervised image captioning often grapples with challenges such as image–text mismatches and modality gaps, resulting in suboptimal captions. This paper introduces a semantic-enhanced cross-modal fusion model (SCFM) to address these issues. The SCFM integrates three innovative components: a text semantic enhancement network (TSE-Net) for nuanced semantic representation; contrast learning for optimizing similarity measures between text and images; and enhanced visual selection decoding (EVSD) for precise captioning. Unlike existing methods that struggle with capturing accurate semantic relationships and flexibility across scenarios, the proposed model provides a robust solution for unbiased and diverse captioning. Through experimental evaluations on the MS COCO and Flickr30k datasets, SCFM demonstrates significant improvements over the benchmark model, enhancing the CIDEr and BLEU-4 metrics by 3.6% and 3.2%, respectively. Visualization analysis further reveals the model’s superiority in increasing variability between hidden features and its potential in cross-domain and stylized image captioning. The findings not only contribute to the advancement of image captioning techniques but also open avenues for future research. Further investigations will explore SCFM’s adaptability to other multimodal tasks and refine it for more intricate image–text relationships.

Both accuracy and efficiency are crucial for image captioning in real-world scenarios. Although Transformer-based models have gained significant improved captioning performance, their computational cost is very high. A feasible way to reduce the time complexity is to exit the prediction early in internal decoding layers without passing the entire model. However, it is not straightforward to devise early exiting into image captioning due to the following issues. On one hand, the representation in shallow layers lacks high-level semantic and sufficient cross-modal fusion information for accurate prediction. On the other hand, the exiting decisions made by internal classifiers are unreliable sometimes. To solve these issues, we propose DeeCap framework for efficient image captioning, which dynamically selects proper-sized decoding layers from a global perspective to exit early. The key to successful early exiting lies in the specially designed imitation learning mechanism, which predicts the deep layer activation with shallow layer features. By deliberately merging the imitation learning into the whole image captioning architecture, the imitated deep layer representation can mitigate the loss brought by the missing of actual deep layers when early exiting is undertaken, resulting in significant reduction in calculation cost with small sacrifice of accuracy. Experiments on the MS COCO and Flickr30k datasets demonstrate the DeeCap can achieve competitive performances with 4× speed-up. Code is available at: https://github.com/feizc/DeeCap.

Image Captioning is the process of generating textual descriptions of an image, which need to be syntactically and semantically correct. This paper extensively surveys very early literature that includes the advent of Artificial Intelligence, the Machine Learning pathway, the early Deep Learning and the current Deep Learning methodology for Image Captioning. This survey paper aims to develop a system to predict captions for the given images with a higher accuracy by combining the results of different Deep Learning Techniques. This model based on a neural network consists of a vision CNN followed by the language generator RNN. It generates complete sentence in natural language from an input image. The state of art is achieved by comparing three different encoder –decoder models. By comparing three models, the blue score of CNN-LSTM Model with Flikr 8k dataset is 0.44, CNN-LSTM Word Embedding with Flikr 8k dataset is 0.68 and CNN –GRU model Visual Attention with MSCOCO Dataset is 0.86.

Image captioning aims at understanding various semantic concepts (e.g., objects and relationships) from an image and integrating them in a sentence-level description. Hence, it is necessary to learn the interaction among these concepts. If we define the context of the interaction to be involved in the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">subject-predicate-object</i> triplet, most current methods only focus on the single triplet for the first-order interaction to generate sentences. Intuitively, we humans are able to perceive the high-order interaction among concepts from two or more triplets to describe an image. For example, when we see the triplets <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">man-cutting-sandwich</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">man-with-knife</i> , it is natural to integrate and predict the sentence <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">man cutting sandwich with knife</i> . This depends on the high-order interaction between <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">cutting</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">knife</i> in different triplets. Therefore, exploiting high-order interaction is expected to benefit image captioning and focus on reasoning. In this paper, we introduce the novel high-order interaction learning method over detected objects and relationships for image captioning under the umbrella of the encoder-decoder framework. We first extract a set of object and relationship features in an image. During the encoding stage, the interactive refining network is proposed to learn high-order representations by modeling intra- and inter-object feature interaction in the self-attention fashion. During the decoding stage, the interactive fusion network is proposed to integrate object and relationship information by strengthening their high-order interaction based on language context for sentence generation. In this way, we learn the object-relationship dependencies in different stages, which can provide abundant cues for both visual understanding and caption generation. Extensive experiments show that the proposed method can achieve competitive performances against the state-of-the-art methods on MSCOCO dataset. Additional ablation studies further validate its effectiveness.

Image captioning is the process of generating a textual description of an image that aims to describe the salient parts of the given image. It is an important problem, as it involves computer vision and natural language processing, where computer vision is used for understanding images, and natural language processing is used for language modeling. A lot of works have been done for image captioning for the English language. In this article, we have developed a model for image captioning in the Hindi language. Hindi is the official language of India, and it is the fourth most spoken language in the world, spoken in India and South Asia. To the best of our knowledge, this is the first attempt to generate image captions in the Hindi language. A dataset is manually created by translating well known MSCOCO dataset from English to Hindi. Finally, different types of attention-based architectures are developed for image captioning in the Hindi language. These attention mechanisms are new for the Hindi language, as those have never been used for the Hindi language. The obtained results of the proposed model are compared with several baselines in terms of BLEU scores, and the results show that our model performs better than others. Manual evaluation of the obtained captions in terms of adequacy and fluency also reveals the effectiveness of our proposed approach. Availability of resources : The codes of the article are available at https://github.com/santosh1821cs03/Image_Captioning_Hindi_Language ; The dataset will be made available: http://www.iitp.ac.in/∼ai-nlp-ml/resources.html .

Describing an image with a grammatically and semantically correct sentence, known as image captioning, has been improved significantly with recent advances in computer vision (CV) and natural language processing (NLP) communities. The integration of these communities leads to the development of feature-injection architectures, which define how extracted features are used in captioning. In this paper, a benchmark of feature-injection architectures that utilize CV and NLP techniques is reported for encoder-decoder based captioning. Benchmark evaluations include Inception-v3 convolutional neural network to extract image features in the encoder while the feature-injection architectures such as init-inject, pre-inject, par-inject and merge are applied with a multi-layer gated recurrent unit (GRU) to generate captions in the decoder. Architectures have been evaluated extensively on the MSCOCO dataset across eight performance metrics. It has been concluded that the init-inject architecture with 3-layer GRU outperforms the other architectures in terms of captioning accuracy.

Recently, the transformer model has been utilized for a variety of tasks such as image captioning, image classification natural language generation, and natural language understanding. As a key component of the transformer model, self-attention calculates the attention values by mapping the relationships among the head elements of the source and target sequence, yet there is no explicit mechanism to refine and intensify the attention values with respect to the context of the input and target sequences. Based on this intuition, we introduce a novel refine and intensify attention mechanism that is called Zoneup Dropout Injection Attention Calculation (ZoDIAC), in which the intensities of attention values in the elements of the input source and target sequences are first refined using GELU and dropout and then intensified using a proposed zoneup process which includes the injection of a learned scalar factor. Our extensive experiments show that ZoDIAC achieves statistically significant higher scores under all image captioning metrics. The code for our experiments is publicly available at: https://github.com/zanyarz/zodiac .

Parallel-fusion LSTM with synchronous semantic and visual information for image captioning

Visual attention has been successfully applied in image captioning to selectively incorporate the most relevant areas to the language generation procedure. However, the attention in current image captioning methods is only guided by the hidden state of language model, e.g. LSTM (Long-Short Term Memory), indirectly and implicitly, and thus the attended areas are weakly relevant at different time steps. Besides the spatial relationship of attention areas, the temporal relationship in attention is crucial for image captioning according to the attention transmission mechanism of human vision. In this paper, we propose a new spatio-temporal memory attention (STMA) model to learn the spatio-temporal relationship in attention for image captioning. The STMA introduces the memory mechanism to the attention model through a tailored LSTM, where the new cell is used to memorize and propagate the attention information, and the output gate is used to generate attention weights. The attention in STMA transmits with memory adaptively and dependently, which builds strong temporal connections of attentions and learns the spatio-temporal relationship of attended areas simultaneously. Besides, the proposed STMA is flexible to combine with attention-based image captioning frameworks. Experiments on MS COCO dataset demonstrate the superiority of the proposed STMA model in exploring the spatio-temporal relationship in attention and improving the current attention-based image captioning.