Fangxiang Feng

Zebin Yao, Fangxiang Feng, Ruifan Li et al.

The customization of text-to-image models has seen significant advancements, yet generating multiple personalized concepts remains a challenging task. Current methods struggle with attribute leakage and layout confusion when handling multiple concepts, leading to reduced concept fidelity and semantic consistency. In this work, we introduce a novel training-free framework, Concept Conductor, designed to ensure visual fidelity and correct layout in multi-concept customization. Concept Conductor isolates the sampling processes of multiple custom models to prevent attribute leakage between different concepts and corrects erroneous layouts through self-attention-based spatial guidance. Additionally, we present a concept injection technique that employs shape-aware masks to specify the generation area for each concept. This technique injects the structure and appearance of personalized concepts through feature fusion in the attention layers, ensuring harmony in the final image. Extensive qualitative and quantitative experiments demonstrate that Concept Conductor can consistently generate composite images with accurate layouts while preserving the visual details of each concept. Compared to existing baselines, Concept Conductor shows significant performance improvements. Our method supports the combination of any number of concepts and maintains high fidelity even when dealing with visually similar concepts. The code and models are available at https://github.com/Nihukat/Concept-Conductor.

Duo Zheng, Fandong Meng, Qingyi Si et al. · tsinghua

Visual dialog has witnessed great progress after introducing various vision-oriented goals into the conversation, especially such as GuessWhich and GuessWhat, where the only image is visible by either and both of the questioner and the answerer, respectively. Researchers explore more on visual dialog tasks in such kind of single- or perfectly co-observable visual scene, while somewhat neglect the exploration on tasks of non perfectly co-observable visual scene, where the images accessed by two agents may not be exactly the same, often occurred in practice. Although building common ground in non-perfectly co-observable visual scene through conversation is significant for advanced dialog agents, the lack of such dialog task and corresponding large-scale dataset makes it impossible to carry out in-depth research. To break this limitation, we propose an object-referring game in non-perfectly co-observable visual scene, where the goal is to spot the difference between the similar visual scenes through conversing in natural language. The task addresses challenges of the dialog strategy in non-perfectly co-observable visual scene and the ability of categorizing objects. Correspondingly, we construct a large-scale multimodal dataset, named SpotDiff, which contains 87k Virtual Reality images and 97k dialogs generated by self-play. Finally, we give benchmark models for this task, and conduct extensive experiments to evaluate its performance as well as analyze its main challenges.

Ruonan Wang, Yuxi Qian, Fangxiang Feng et al.

Most existing approaches to Visual Question Answering (VQA) answer questions directly, however, people usually decompose a complex question into a sequence of simple sub questions and finally obtain the answer to the original question after answering the sub question sequence(SQS). By simulating the process, this paper proposes a conversation-based VQA (Co-VQA) framework, which consists of three components: Questioner, Oracle, and Answerer. Questioner raises the sub questions using an extending HRED model, and Oracle answers them one-by-one. An Adaptive Chain Visual Reasoning Model (ACVRM) for Answerer is also proposed, where the question-answer pair is used to update the visual representation sequentially. To perform supervised learning for each model, we introduce a well-designed method to build a SQS for each question on VQA 2.0 and VQA-CP v2 datasets. Experimental results show that our method achieves state-of-the-art on VQA-CP v2. Further analyses show that SQSs help build direct semantic connections between questions and images, provide question-adaptive variable-length reasoning chains, and with explicit interpretability as well as error traceability.

Yuxi Qian, Yuncong Hu, Ruonan Wang et al.

Existing Visual Question Answering (VQA) models have explored various visual relationships between objects in the image to answer complex questions, which inevitably introduces irrelevant information brought by inaccurate object detection and text grounding. To address the problem, we propose a Question-Driven Graph Fusion Network (QD-GFN). It first models semantic, spatial, and implicit visual relations in images by three graph attention networks, then question information is utilized to guide the aggregation process of the three graphs, further, our QD-GFN adopts an object filtering mechanism to remove question-irrelevant objects contained in the image. Experiment results demonstrate that our QD-GFN outperforms the prior state-of-the-art on both VQA 2.0 and VQA-CP v2 datasets. Further analysis shows that both the novel graph aggregation method and object filtering mechanism play a significant role in improving the performance of the model.

Fulong Ye, Yuxing Long, Fangxiang Feng et al.

Referring Expression Generation (REG) aims to generate unambiguous Referring Expressions (REs) for objects in a visual scene, with a dual task of Referring Expression Comprehension (REC) to locate the referred object. Existing methods construct REG models independently by using only the REs as ground truth for model training, without considering the potential interaction between REG and REC models. In this paper, we propose an Interactive REG (IREG) model that can interact with a real REC model, utilizing signals indicating whether the object is located and the visual region located by the REC model to gradually modify REs. Our experimental results on three RE benchmark datasets, RefCOCO, RefCOCO+, and RefCOCOg show that IREG outperforms previous state-of-the-art methods on popular evaluation metrics. Furthermore, a human evaluation shows that IREG generates better REs with the capability of interaction.

Bo Yang, Fangxiang Feng, Xiaojie Wang

A good Text-to-Image model should not only generate high quality images, but also ensure the consistency between the text and the generated image. Previous models failed to simultaneously fix both sides well. This paper proposes a Gradual Refinement Generative Adversarial Network (GR-GAN) to alleviates the problem efficiently. A GRG module is designed to generate images from low resolution to high resolution with the corresponding text constraints from coarse granularity (sentence) to fine granularity (word) stage by stage, a ITM module is designed to provide image-text matching losses at both sentence-image level and word-region level for corresponding stages. We also introduce a new metric Cross-Model Distance (CMD) for simultaneously evaluating image quality and image-text consistency. Experimental results show GR-GAN significant outperform previous models, and achieve new state-of-the-art on both FID and CMD. A detailed analysis demonstrates the efficiency of different generation stages in GR-GAN.

Shuhao Gu, Jialing Zhang, Siyuan Zhou et al.

Recently, Vision-Language Models (VLMs) have achieved remarkable progress in multimodal tasks, and multimodal instruction data serves as the foundation for enhancing VLM capabilities. Despite the availability of several open-source multimodal datasets, limitations in the scale and quality of open-source instruction data hinder the performance of VLMs trained on these datasets, leading to a significant gap compared to models trained on closed-source data. To address this challenge, we introduce Infinity-MM, a large-scale multimodal instruction dataset. We collected the available multimodal instruction datasets and performed unified preprocessing, resulting in a dataset with over 40 million samples that ensures diversity and accuracy. Furthermore, to enable large-scale expansion of instruction data and support the continuous acquisition of high-quality data, we propose a synthetic instruction generation method based on a tagging system and open-source VLMs. By establishing correspondences between different types of images and associated instruction types, this method can provide essential guidance during data synthesis. Leveraging this high-quality data, we have trained a 2-billion-parameter Vision-Language Model, Aquila-VL-2B, which achieves state-of-the-art (SOTA) performance among models of similar scale. The data is available at: https://huggingface.co/datasets/BAAI/Infinity-MM.

Pengfei Zhou, Fangxiang Feng, Xiaojie Wang

Image harmonization, which involves adjusting the foreground of a composite image to attain a unified visual consistency with the background, can be conceptualized as an image-to-image translation task. Diffusion models have recently promoted the rapid development of image-to-image translation tasks . However, training diffusion models from scratch is computationally intensive. Fine-tuning pre-trained latent diffusion models entails dealing with the reconstruction error induced by the image compression autoencoder, making it unsuitable for image generation tasks that involve pixel-level evaluation metrics. To deal with these issues, in this paper, we first adapt a pre-trained latent diffusion model to the image harmonization task to generate the harmonious but potentially blurry initial images. Then we implement two strategies: utilizing higher-resolution images during inference and incorporating an additional refinement stage, to further enhance the clarity of the initially harmonized images. Extensive experiments on iHarmony4 datasets demonstrate the superiority of our proposed method. The code and model will be made publicly available at https://github.com/nicecv/DiffHarmony .

Ziming Cheng, Binrui Xu, Lisheng Gong et al.

With enhanced capabilities and widespread applications, Multimodal Large Language Models (MLLMs) are increasingly required to process and reason over multiple images simultaneously. However, existing MLLM benchmarks focus either on single-image visual reasoning or on multi-image understanding tasks with only final-answer evaluation, leaving the reasoning capabilities of MLLMs over multi-image inputs largely underexplored. To address this gap, we introduce the $\textbf{Multimodal Multi-image Reasoning Benchmark (MMRB)}$, the first benchmark designed to evaluate structured visual reasoning across multiple images. MMRB comprises $\textbf{92 sub-tasks}$ covering spatial, temporal, and semantic reasoning, with multi-solution, CoT-style annotations generated by GPT-4o and refined by human experts. A derivative subset is designed to evaluate multimodal reward models in multi-image scenarios. To support fast and scalable evaluation, we propose a sentence-level matching framework using open-source LLMs. Extensive baseline experiments on $\textbf{40 MLLMs}$, including 9 reasoning-specific models and 8 reward models, demonstrate that open-source MLLMs still lag significantly behind commercial MLLMs in multi-image reasoning tasks. Furthermore, current multimodal reward models are nearly incapable of handling multi-image reward ranking tasks.

Zebin Yao, Lei Ren, Huixing Jiang et al.

Subject-driven image generation aims to synthesize novel scenes that faithfully preserve subject identity from reference images while adhering to textual guidance. However, existing methods struggle with a critical trade-off between fidelity and efficiency. Tuning-based approaches rely on time-consuming and resource-intensive, subject-specific optimization, while zero-shot methods often fail to maintain adequate subject consistency. In this work, we propose FreeGraftor, a training-free framework that addresses these limitations through cross-image feature grafting. Specifically, FreeGraftor leverages semantic matching and position-constrained attention fusion to transfer visual details from reference subjects to the generated images. Additionally, our framework introduces a novel noise initialization strategy to preserve the geometry priors of reference subjects, facilitating robust feature matching. Extensive qualitative and quantitative experiments demonstrate that our method enables precise subject identity transfer while maintaining text-aligned scene synthesis. Without requiring model fine-tuning or additional training, FreeGraftor significantly outperforms existing zero-shot and training-free approaches in both subject fidelity and text alignment. Furthermore, our framework can seamlessly extend to multi-subject generation, making it practical for real-world deployment. Our code is available at https://github.com/Nihukat/FreeGraftor.

Guoyang Xia, Yifeng Ding, Fengfa Li et al.

Multimodal large language models (MLLMs) have achieved impressive performance, but high-resolution visual inputs result in long sequences of visual tokens and substantial inference latency. Reducing redundant visual tokens is critical to ease computational/memory burdens while preserving performance, enabling MLLM deployment in resource-constrained or latency-sensitive scenarios. Current visual token pruning methods mainly rely on attention-based redundancy analysis and are tailored to dense architectures. We propose Fast Multimodal Mixture-of-Experts (FastMMoE), a training-free acceleration framework for mixture-of-experts (MoE) based MLLMs, developed from a routing analysis perspective. FastMMoE combines two complementary strategies: (i) expert activation reduction for visual tokens to minimize unnecessary expert computation; and (ii) routing-aware token pruning that leverages similarity in routing probability distributions to identify and remove highly redundant visual tokens. Experiments on large-scale MoE-MLLMs such as DeepSeek-VL2 and InternVL3.5 demonstrate that FastMMoE can reduce FLOPs by up to 55.0% while retaining approximately 95.5% of the original performance, consistently outperforming dense-model pruning baselines including FastV and SparseVLM across multiple retention rates.

Haojie Ouyang, Jianwei Lv, Lei Ren et al.

Transformer-based large models excel in natural language processing and computer vision, but face severe computational inefficiencies due to the self-attention's quadratic complexity with input tokens. Recently, researchers have proposed a series of methods based on block selection and compression to alleviate this problem, but they either have issues with semantic incompleteness or poor training-inference efficiency. To comprehensively address these challenges, we propose ChunkLLM, a lightweight and pluggable training framework. Specifically, we introduce two components: QK Adapter (Q-Adapter and K-Adapter) and Chunk Adapter. The former is attached to each Transformer layer, serving dual purposes of feature compression and chunk attention acquisition. The latter operates at the bottommost layer of the model, functioning to detect chunk boundaries by leveraging contextual semantic information. During the training phase, the parameters of the backbone remain frozen, with only the QK Adapter and Chunk Adapter undergoing training. Notably, we design an attention distillation method for training the QK Adapter, which enhances the recall rate of key chunks. During the inference phase, chunk selection is triggered exclusively when the current token is detected as a chunk boundary, thereby accelerating model inference. Experimental evaluations are conducted on a diverse set of long-text and short-text benchmark datasets spanning multiple tasks. ChunkLLM not only attains comparable performance on short-text benchmarks but also maintains 98.64% of the performance on long-context benchmarks while preserving a 48.58% key-value cache retention rate. Particularly, ChunkLLM attains a maximum speedup of 4.48x in comparison to the vanilla Transformer in the processing of 120K long texts.

Tongtong Liu, Fangxiang Feng, Xiaojie Wang

Multimodal pre-training models, such as LXMERT, have achieved excellent results in downstream tasks. However, current pre-trained models require large amounts of training data and have huge model sizes, which make them difficult to apply in low-resource situations. How to obtain similar or even better performance than a larger model under the premise of less pre-training data and smaller model size has become an important problem. In this paper, we propose a new Multi-stage Pre-training (MSP) method, which uses information at different granularities from word, phrase to sentence in both texts and images to pre-train the model in stages. We also design several different pre-training tasks suitable for the information granularity in different stage in order to efficiently capture the diverse knowledge from a limited corpus. We take a Simplified LXMERT (LXMERT- S), which has only 45.9% parameters of the original LXMERT model and 11.76% of the original pre-training data as the testbed of our MSP method. Experimental results show that our method achieves comparable performance to the original LXMERT model in all downstream tasks, and even outperforms the original model in Image-Text Retrieval task.

Zipeng Xu, Fangxiang Feng, Xiaojie Wang et al.

A goal-oriented visual dialogue involves multi-turn interactions between two agents, Questioner and Oracle. During which, the answer given by Oracle is of great significance, as it provides golden response to what Questioner concerns. Based on the answer, Questioner updates its belief on target visual content and further raises another question. Notably, different answers drive into different visual beliefs and future questions. However, existing methods always indiscriminately encode answers after much longer questions, resulting in a weak utilization of answers. In this paper, we propose an Answer-Driven Visual State Estimator (ADVSE) to impose the effects of different answers on visual states. First, we propose an Answer-Driven Focusing Attention (ADFA) to capture the answer-driven effect on visual attention by sharpening question-related attention and adjusting it by answer-based logical operation at each turn. Then based on the focusing attention, we get the visual state estimation by Conditional Visual Information Fusion (CVIF), where overall information and difference information are fused conditioning on the question-answer state. We evaluate the proposed ADVSE to both question generator and guesser tasks on the large-scale GuessWhat?! dataset and achieve the state-of-the-art performances on both tasks. The qualitative results indicate that the ADVSE boosts the agent to generate highly efficient questions and obtains reliable visual attentions during the reasonable question generation and guess processes.

Fangxiang Feng, Ruifan Li, Xiaojie Wang

This paper describes our solution to the multi-modal learning challenge of ICML. This solution comprises constructing three-level representations in three consecutive stages and choosing correct tag words with a data-specific strategy. Firstly, we use typical methods to obtain level-1 representations. Each image is represented using MPEG-7 and gist descriptors with additional features released by the contest organizers. And the corresponding word tags are represented by bag-of-words model with a dictionary of 4000 words. Secondly, we learn the level-2 representations using two stacked RBMs for each modality. Thirdly, we propose a bimodal auto-encoder to learn the similarities/dissimilarities between the pairwise image-tags as level-3 representations. Finally, during the test phase, based on one observation of the dataset, we come up with a data-specific strategy to choose the correct tag words leading to a leap of an improved overall performance. Our final average accuracy on the private test set is 100%, which ranks the first place in this challenge.

Ian J. Goodfellow, Dumitru Erhan, Pierre Luc Carrier et al.

The ICML 2013 Workshop on Challenges in Representation Learning focused on three challenges: the black box learning challenge, the facial expression recognition challenge, and the multimodal learning challenge. We describe the datasets created for these challenges and summarize the results of the competitions. We provide suggestions for organizers of future challenges and some comments on what kind of knowledge can be gained from machine learning competitions.