Zengchang Qin

Yuhe Liu, Chuanjian Liu, Kai Han et al.

Semantic segmentation is a computer vision task that associates a label with each pixel in an image. Modern approaches tend to introduce class embeddings into semantic segmentation for deeply utilizing category semantics, and regard supervised class masks as final predictions. In this paper, we explore the mechanism of class embeddings and have an insight that more explicit and meaningful class embeddings can be generated based on class masks purposely. Following this observation, we propose ECENet, a new segmentation paradigm, in which class embeddings are obtained and enhanced explicitly during interacting with multi-stage image features. Based on this, we revisit the traditional decoding process and explore inverted information flow between segmentation masks and class embeddings. Furthermore, to ensure the discriminability and informativity of features from backbone, we propose a Feature Reconstruction module, which combines intrinsic and diverse branches together to ensure the concurrence of diversity and redundancy in features. Experiments show that our ECENet outperforms its counterparts on the ADE20K dataset with much less computational cost and achieves new state-of-the-art results on PASCAL-Context dataset. The code will be released at https://gitee.com/mindspore/models and https://github.com/Carol-lyh/ECENet.

Yubo Hou, Jingwei Song, Hongbo Zhang et al.

Persistent LLM agents require memory representations that make the formation of person understanding explicit across long term interaction. Existing agent memory methods emphasize information retention and retrieval, yet give limited account of how accumulated interaction evidence is abstracted into person understanding. We view this process as schema formation, where situated evidence is abstracted into reusable patterns and stable person level claims. We introduce PersonaTree, a structured lifecycle memory framework that realizes this view as a three level persona tree with explicit support paths from evidence to claims. PersonaTree maintains the tree through conservative writing, confidence guided consolidation, and query conditioned path retrieval, returning only the evidence depth required by each query. Across six person understanding and persistent memory benchmarks with three answer backbones, PersonaTree ranks first in 12 of 18 compact scores and reaches the top two in 16 settings. Ablations show that hierarchy improves abstract person understanding on KnowMe, while support path retrieval improves RealPref alignment under a comparable context budget.

Zheng He, Zeke Xie, Quanzhi Zhu et al.

People usually believe that network pruning not only reduces the computational cost of deep networks, but also prevents overfitting by decreasing model capacity. However, our work surprisingly discovers that network pruning sometimes even aggravates overfitting. We report an unexpected sparse double descent phenomenon that, as we increase model sparsity via network pruning, test performance first gets worse (due to overfitting), then gets better (due to relieved overfitting), and gets worse at last (due to forgetting useful information). While recent studies focused on the deep double descent with respect to model overparameterization, they failed to recognize that sparsity may also cause double descent. In this paper, we have three main contributions. First, we report the novel sparse double descent phenomenon through extensive experiments. Second, for this phenomenon, we propose a novel learning distance interpretation that the curve of $\ell_{2}$ learning distance of sparse models (from initialized parameters to final parameters) may correlate with the sparse double descent curve well and reflect generalization better than minima flatness. Third, in the context of sparse double descent, a winning ticket in the lottery ticket hypothesis surprisingly may not always win.

Shunyu Zhang, Xiaoze Jiang, Zequn Yang et al.

Visual Dialog requires an agent to engage in a conversation with humans grounded in an image. Many studies on Visual Dialog focus on the understanding of the dialog history or the content of an image, while a considerable amount of commonsense-required questions are ignored. Handling these scenarios depends on logical reasoning that requires commonsense priors. How to capture relevant commonsense knowledge complementary to the history and the image remains a key challenge. In this paper, we propose a novel model by Reasoning with Multi-structure Commonsense Knowledge (RMK). In our model, the external knowledge is represented with sentence-level facts and graph-level facts, to properly suit the scenario of the composite of dialog history and image. On top of these multi-structure representations, our model can capture relevant knowledge and incorporate them into the vision and semantic features, via graph-based interaction and transformer-based fusion. Experimental results and analysis on VisDial v1.0 and VisDialCK datasets show that our proposed model effectively outperforms comparative methods.

Yubo Hou, Zhisheng Chen, Tao Wan et al.

The stateless architecture of Large Language Models inherently lacks the mechanism to preserve dynamic context, compelling agents to redundantly reprocess history to maintain long-horizon autonomy. While latent memory offers a solution, current approaches are hindered by architectural segregation, relying on auxiliary encoders that decouple memory from the reasoning backbone. We propose FlashMem, a framework that distills intrinsic memory directly from transient reasoning states via computation reuse. Leveraging the property that internal representations uniquely encode input trajectories, FlashMem identifies the last hidden state as a sufficient statistic for the interaction history. This enables a Shared-KV Consolidator to synthesize memory by attending directly to the backbone's frozen cache, eliminating redundant re-parameterization. Furthermore, a parameter-free Cognitive Monitor leverages attention entropy to adaptively trigger consolidation only when high epistemic uncertainty is detected. Experiments demonstrate that FlashMem matches the performance of heavy baselines while reducing inference latency by 5 times, effectively bridging the gap between efficiency and persistent cognition.

Qingyun Jin, Xiaohui Song, Feng Zhou et al.

Large language models (LLMs) have demonstrated exceptional performance across diverse natural language processing tasks. However, as the model size and the input sequence's length increase, the linearly increasing key-value (KV) cache significantly degrades inference throughput. Therefore, grouped-query attention (GQA), as an alternative to multi-head attention (MHA), has been widely introduced into LLMs. In this work, we propose a cost-effective method for converting MHA into GQA with any compression ratio of KV heads. The key point of our method lies in the application of Procrustes analysis to the attention heads, which enhances the similarity among attention heads while preserving computational invariance, thereby improving the model's post-training performance. Subsequently, we employ $\mathit{L_0}$ regularization to prune redundant parameters. The model after pruning can be adapted to the standard GQA framework. Experimental results show that our strategy can compress up to 87.5\% KV heads of LLaMA2-7B model and 75\% KV heads of Sheared-LLaMA-1.3B with acceptable performance degradation. Our code is released at https://github.com/fpcsong/mha2gqa.

Ziije Zhong, Linqing Zhong, Zhaoze Sun et al.

Integrating Large Language Models (LLMs) with existing Knowledge Graph (KG) databases presents a promising avenue for enhancing LLMs' efficacy and mitigating their "hallucinations". Given that most KGs reside in graph databases accessible solely through specialized query languages (e.g., Cypher), it is critical to connect LLMs with KG databases by automating the translation of natural language into Cypher queries (termed as "Text2Cypher" task). Prior efforts tried to bolster LLMs' proficiency in Cypher generation through Supervised Fine-Tuning (SFT). However, these explorations are hindered by the lack of annotated datasets of Query-Cypher pairs, resulting from the labor-intensive and domain-specific nature of such annotation. In this study, we propose SyntheT2C, a methodology for constructing a synthetic Query-Cypher pair dataset, comprising two distinct pipelines: (1) LLM-based prompting and (2) template-filling. SyntheT2C is applied to two medical KG databases, culminating in the creation of a synthetic dataset, MedT2C. Comprehensive experiments demonstrate that the MedT2C dataset effectively enhances the performance of backbone LLMs on Text2Cypher task via SFT. Both the SyntheT2C codebase and the MedT2C dataset are released in https://github.com/ZGChung/SyntheT2C.

Zijie Zhong, Hanwen Liu, Xiaoya Cui et al.

Integrating information from various reference databases is a major challenge for Retrieval-Augmented Generation (RAG) systems because each knowledge source adopts a unique data structure and follows different conventions. Retrieving from multiple knowledge sources with one fixed strategy usually leads to under-exploitation of information. To mitigate this drawback, inspired by Mix-of-Expert, we introduce Mix-of-Granularity (MoG), a method that dynamically determines the optimal granularity of a knowledge source based on input queries using a router. The router is efficiently trained with a newly proposed loss function employing soft labels. We further extend MoG to MoG-Graph (MoGG), where reference documents are pre-processed as graphs, enabling the retrieval of distantly situated snippets. Experiments demonstrate that MoG and MoGG effectively predict optimal granularity levels, significantly enhancing the performance of the RAG system in downstream tasks. The code of both MoG and MoGG are released in https://github.com/ZGChung/Mix-of-Granularity.

Zijie Zhong, Yunhui Zhang, Ziyi Chang et al.

Node Importance Estimation (NIE) is crucial for integrating external information into Large Language Models through Retriever-Augmented Generation. Traditional methods, focusing on static, single-graph characteristics, lack adaptability to new graphs and user-specific requirements. CADReN, our proposed method, addresses these limitations by introducing a Contextual Anchor (CA) mechanism. This approach enables the network to assess node importance relative to the CA, considering both structural and semantic features within Knowledge Graphs (KGs). Extensive experiments show that CADReN achieves better performance in cross-graph NIE task, with zero-shot prediction ability. CADReN is also proven to match the performance of previous models on single-graph NIE task. Additionally, we introduce and opensource two new datasets, RIC200 and WK1K, specifically designed for cross-graph NIE research, providing a valuable resource for future developments in this domain.

Xiaoze Jiang, Jing Yu, Yajing Sun et al.

Visual Dialogue task requires an agent to be engaged in a conversation with human about an image. The ability of generating detailed and non-repetitive responses is crucial for the agent to achieve human-like conversation. In this paper, we propose a novel generative decoding architecture to generate high-quality responses, which moves away from decoding the whole encoded semantics towards the design that advocates both transparency and flexibility. In this architecture, word generation is decomposed into a series of attention-based information selection steps, performed by the novel recurrent Deliberation, Abandon and Memory (DAM) module. Each DAM module performs an adaptive combination of the response-level semantics captured from the encoder and the word-level semantics specifically selected for generating each word. Therefore, the responses contain more detailed and non-repetitive descriptions while maintaining the semantic accuracy. Furthermore, DAM is flexible to cooperate with existing visual dialogue encoders and adaptive to the encoder structures by constraining the information selection mode in DAM. We apply DAM to three typical encoders and verify the performance on the VisDial v1.0 dataset. Experimental results show that the proposed models achieve new state-of-the-art performance with high-quality responses. The code is available at https://github.com/JXZe/DAM.

Suyuan Huang, Haoxin Zhang, Linqing Zhong et al.

Covering from Image LLMs to the more complex Video LLMs, the Multimodal Large Language Models (MLLMs) have demonstrated profound capabilities in comprehending cross-modal information as numerous studies have illustrated. Previous methods delve into designing comprehensive Video LLMs through integrating video foundation models with primitive LLMs. Despite its effectiveness, such paradigm renders Video LLM's structure verbose and typically requires substantial video data for pre-training. Crucially, it neglects leveraging the foundational contributions of ready-made Image LLMs. In this paper, we introduce RED-VILLM, a Resource-Efficient Development pipeline which builds robust Video LLMs through leveraging the prior knowledge of Image LLMs. Specifically, since a video is naturally a combination of images along the temporal dimension, we devise a temporal adaptation plug-and-play structure, endowing the backbone Image LLM with the capability to grasp temporal information. Moreover, through applying this pipeline, we achieve the first Video LLM within the Chinese-speaking community. Extensive experiments demonstrate that Video LLMs developed through our approach surpass conventional Video LLMs, requiring minimal instructional data and training resources. Our approach highlights the potential for a more cost-effective and scalable advancement in multimodal models.

Qianhao Han, Junyi Liu, Zengchang Qin et al.

Automating radiology report generation can significantly reduce the workload of radiologists and enhance the accuracy, consistency, and efficiency of clinical documentation.We propose a novel cross-modal framework that uses MedCLIP as both a vision extractor and a retrieval mechanism to improve the process of medical report generation.By extracting retrieved report features and image features through an attention-based extract module, and integrating them with a fusion module, our method improves the coherence and clinical relevance of generated reports.Experimental results on the widely used IU-Xray dataset demonstrate the effectiveness of our approach, showing improvements over commonly used methods in both report quality and relevance.Additionally, ablation studies provide further validation of the framework, highlighting the importance of accurate report retrieval and feature integration in generating comprehensive medical reports.

Yuhe Liu, Mengxue Kang, Zengchang Qin et al.

Large text-to-image models have achieved astonishing performance in synthesizing diverse and high-quality images guided by texts. With detail-oriented conditioning control, even finer-grained spatial control can be achieved. However, some generated images still appear unreasonable, even with plentiful object features and a harmonious style. In this paper, we delve into the underlying causes and find that deep-level logical information, serving as common-sense knowledge, plays a significant role in understanding and processing images. Nonetheless, almost all models have neglected the importance of logical relations in images, resulting in poor performance in this aspect. Following this observation, we propose LogicalDefender, which combines images with the logical knowledge already summarized by humans in text. This encourages models to learn logical knowledge faster and better, and concurrently, extracts the widely applicable logical knowledge from both images and human knowledge. Experiments show that our model has achieved better logical performance, and the extracted logical knowledge can be effectively applied to other scenarios.

Xiaoze Jiang, Siyi Du, Zengchang Qin et al.

Visual dialogue is a challenging task that needs to extract implicit information from both visual (image) and textual (dialogue history) contexts. Classical approaches pay more attention to the integration of the current question, vision knowledge and text knowledge, despising the heterogeneous semantic gaps between the cross-modal information. In the meantime, the concatenation operation has become de-facto standard to the cross-modal information fusion, which has a limited ability in information retrieval. In this paper, we propose a novel Knowledge-Bridge Graph Network (KBGN) model by using graph to bridge the cross-modal semantic relations between vision and text knowledge in fine granularity, as well as retrieving required knowledge via an adaptive information selection mode. Moreover, the reasoning clues for visual dialogue can be clearly drawn from intra-modal entities and inter-modal bridges. Experimental results on VisDial v1.0 and VisDial-Q datasets demonstrate that our model outperforms existing models with state-of-the-art results.

Ying Huang, Jiankai Zhuang, Zengchang Qin

In multi-person pose estimation, the left/right joint type discrimination is always a hard problem because of the similar appearance. Traditionally, we solve this problem by stacking multiple refinement modules to increase network's receptive fields and capture more global context, which can also increase a great amount of computation. In this paper, we propose a Multi-level Network (MLN) that learns to aggregate features from lower-level (left/right information), upper-level (localization information), joint-limb level (complementary information) and global-level (context) information for discrimination of joint type. Through feature reuse and its intra-relation, MLN can attain comparable performance to other conventional methods while runtime speed retains at 42.2 FPS.

Ying Huang, Bin Sun, Haipeng Kan et al.

Human pose estimation has made significant advancement in recent years. However, the existing datasets are limited in their coverage of pose variety. In this paper, we introduce a novel benchmark FollowMeUp Sports that makes an important advance in terms of specific postures, self-occlusion and class balance, a contribution that we feel is required for future development in human body models. This comprehensive dataset was collected using an established taxonomy of over 200 standard workout activities with three different shot angles. The collected videos cover a wider variety of specific workout activities than previous datasets including push-up, squat and body moving near the ground with severe self-occlusion or occluded by some sport equipment and outfits. Given these rich images, we perform a detailed analysis of the leading human pose estimation approaches gaining insights for the success and failures of these methods.

Xiaoze Jiang, Jing Yu, Zengchang Qin et al.

Different from Visual Question Answering task that requires to answer only one question about an image, Visual Dialogue involves multiple questions which cover a broad range of visual content that could be related to any objects, relationships or semantics. The key challenge in Visual Dialogue task is thus to learn a more comprehensive and semantic-rich image representation which may have adaptive attentions on the image for variant questions. In this research, we propose a novel model to depict an image from both visual and semantic perspectives. Specifically, the visual view helps capture the appearance-level information, including objects and their relationships, while the semantic view enables the agent to understand high-level visual semantics from the whole image to the local regions. Futhermore, on top of such multi-view image features, we propose a feature selection framework which is able to adaptively capture question-relevant information hierarchically in fine-grained level. The proposed method achieved state-of-the-art results on benchmark Visual Dialogue datasets. More importantly, we can tell which modality (visual or semantic) has more contribution in answering the current question by visualizing the gate values. It gives us insights in understanding of human cognition in Visual Dialogue.

Zhuoqian Yang, Zengchang Qin, Jing Yu et al.

One of the key issues of Visual Question Answering (VQA) is to reason with semantic clues in the visual content under the guidance of the question, how to model relational semantics still remains as a great challenge. To fully capture visual semantics, we propose to reason over a structured visual representation - scene graph, with embedded objects and inter-object relationships. This shows great benefit over vanilla vector representations and implicit visual relationship learning. Based on existing visual relationship models, we propose a visual relationship encoder that projects visual relationships into a learned deep semantic space constrained by visual context and language priors. Upon the constructed graph, we propose a Scene Graph Convolutional Network (SceneGCN) to jointly reason the object properties and relational semantics for the correct answer. We demonstrate the model's effectiveness and interpretability on the challenging GQA dataset and the classical VQA 2.0 dataset, remarkably achieving state-of-the-art 54.56% accuracy on GQA compared to the existing best model.

Daouda Sow, Zengchang Qin, Mouhamed Niasse et al.

The recent advances of deep learning in both computer vision (CV) and natural language processing (NLP) provide us a new way of understanding semantics, by which we can deal with more challenging tasks such as automatic description generation from natural images. In this challenge, the encoder-decoder framework has achieved promising performance when a convolutional neural network (CNN) is used as image encoder and a recurrent neural network (RNN) as decoder. In this paper, we introduce a sequential guiding network that guides the decoder during word generation. The new model is an extension of the encoder-decoder framework with attention that has an additional guiding long short-term memory (LSTM) and can be trained in an end-to-end manner by using image/descriptions pairs. We validate our approach by conducting extensive experiments on a benchmark dataset, i.e., MS COCO Captions. The proposed model achieves significant improvement comparing to the other state-of-the-art deep learning models.

Shuangting Liu, Jiaqi Zhang, Yuxin Chen et al.

Semantic segmentation is one of the basic topics in computer vision, it aims to assign semantic labels to every pixel of an image. Unbalanced semantic label distribution could have a negative influence on segmentation accuracy. In this paper, we investigate using data augmentation approach to balance the semantic label distribution in order to improve segmentation performance. We propose using generative adversarial networks (GANs) to generate realistic images for improving the performance of semantic segmentation networks. Experimental results show that the proposed method can not only improve segmentation performance on those classes with low accuracy, but also obtain 1.3% to 2.1% increase in average segmentation accuracy. It shows that this augmentation method can boost accuracy and be easily applicable to any other segmentation models.

Jing Yu, Chenghao Yang, Zengchang Qin et al.

Feature modeling of different modalities is a basic problem in current research of cross-modal information retrieval. Existing models typically project texts and images into one embedding space, in which semantically similar information will have a shorter distance. Semantic modeling of textural relationships is notoriously difficult. In this paper, we propose an approach to model texts using a featured graph by integrating multi-view textual relationships including semantic relations, statistical co-occurrence, and prior relations in the knowledge base. A dual-path neural network is adopted to learn multi-modal representations of information and cross-modal similarity measure jointly. We use a Graph Convolutional Network (GCN) for generating relation-aware text representations, and use a Convolutional Neural Network (CNN) with non-linearities for image representations. The cross-modal similarity measure is learned by distance metric learning. Experimental results show that, by leveraging the rich relational semantics in texts, our model can outperform the state-of-the-art models by 3.4% and 6.3% on accuracy on two benchmark datasets.

Jing Yu, Yuhang Lu, Zengchang Qin et al.

Cross-modal information retrieval aims to find heterogeneous data of various modalities from a given query of one modality. The main challenge is to map different modalities into a common semantic space, in which distance between concepts in different modalities can be well modeled. For cross-modal information retrieval between images and texts, existing work mostly uses off-the-shelf Convolutional Neural Network (CNN) for image feature extraction. For texts, word-level features such as bag-of-words or word2vec are employed to build deep learning models to represent texts. Besides word-level semantics, the semantic relations between words are also informative but less explored. In this paper, we model texts by graphs using similarity measure based on word2vec. A dual-path neural network model is proposed for couple feature learning in cross-modal information retrieval. One path utilizes Graph Convolutional Network (GCN) for text modeling based on graph representations. The other path uses a neural network with layers of nonlinearities for image modeling based on off-the-shelf features. The model is trained by a pairwise similarity loss function to maximize the similarity of relevant text-image pairs and minimize the similarity of irrelevant pairs. Experimental results show that the proposed model outperforms the state-of-the-art methods significantly, with 17% improvement on accuracy for the best case.

Heng Wang, Zengchang Qin, Tao Wan

In this paper, we propose a model using generative adversarial net (GAN) to generate realistic text. Instead of using standard GAN, we combine variational autoencoder (VAE) with generative adversarial net. The use of high-level latent random variables is helpful to learn the data distribution and solve the problem that generative adversarial net always emits the similar data. We propose the VGAN model where the generative model is composed of recurrent neural network and VAE. The discriminative model is a convolutional neural network. We train the model via policy gradient. We apply the proposed model to the task of text generation and compare it to other recent neural network based models, such as recurrent neural network language model and SeqGAN. We evaluate the performance of the model by calculating negative log-likelihood and the BLEU score. We conduct experiments on three benchmark datasets, and results show that our model outperforms other previous models.

Xinyue Zhu, Yifan Liu, Zengchang Qin et al.

It is a difficult task to classify images with multiple class labels using only a small number of labeled examples, especially when the label (class) distribution is imbalanced. Emotion classification is such an example of imbalanced label distribution, because some classes of emotions like \emph{disgusted} are relatively rare comparing to other labels like {\it happy or sad}. In this paper, we propose a data augmentation method using generative adversarial networks (GAN). It can complement and complete the data manifold and find better margins between neighboring classes. Specifically, we design a framework with a CNN model as the classifier and a cycle-consistent adversarial networks (CycleGAN) as the generator. In order to avoid gradient vanishing problem, we employ the least-squared loss as adversarial loss. We also propose several evaluation methods on three benchmark datasets to validate GAN's performance. Empirical results show that we can obtain 5%~10% increase in the classification accuracy after employing the GAN-based data augmentation techniques.

Liang Li, Pengyu Li, Yifan Liu et al.

Semantic parsing has emerged as a significant and powerful paradigm for natural language interface and question answering systems. Traditional methods of building a semantic parser rely on high-quality lexicons, hand-crafted grammars and linguistic features which are limited by applied domain or representation. In this paper, we propose a general approach to learn from denotations based on Seq2Seq model augmented with attention mechanism. We encode input sequence into vectors and use dynamic programming to infer candidate logical forms. We utilize the fact that similar utterances should have similar logical forms to help reduce the searching space. Under our learning policy, the Seq2Seq model can learn mappings gradually with noises. Curriculum learning is adopted to make the learning smoother. We test our method on the arithmetic domain which shows our model can successfully infer the correct logical forms and learn the word meanings, compositionality and operation orders simultaneously.

Qiangeng Xu, Zengchang Qin, Tao Wan

How to build a good model for image generation given an abstract concept is a fundamental problem in computer vision. In this paper, we explore a generative model for the task of generating unseen images with desired features. We propose the Generative Cooperative Net (GCN) for image generation. The idea is similar to generative adversarial networks except that the generators and discriminators are trained to work accordingly. Our experiments on hand-written digit generation and facial expression generation show that GCN's two cooperative counterparts (the generator and the classifier) can work together nicely and achieve promising results. We also discovered a usage of such generative model as an data-augmentation tool. Our experiment of applying this method on a recognition task shows that it is very effective comparing to other existing methods. It is easy to set up and could help generate a very large synthesized dataset.

Yifan Liu, Zengchang Qin, Zhenbo Luo et al.

Recently, realistic image generation using deep neural networks has become a hot topic in machine learning and computer vision. Images can be generated at the pixel level by learning from a large collection of images. Learning to generate colorful cartoon images from black-and-white sketches is not only an interesting research problem, but also a potential application in digital entertainment. In this paper, we investigate the sketch-to-image synthesis problem by using conditional generative adversarial networks (cGAN). We propose the auto-painter model which can automatically generate compatible colors for a sketch. The new model is not only capable of painting hand-draw sketch with proper colors, but also allowing users to indicate preferred colors. Experimental results on two sketch datasets show that the auto-painter performs better that existing image-to-image methods.