CMMCoT: Enhancing Complex Multi-Image Comprehension via Multi-Modal Chain-of-Thought and Memory Augmentation

While previous multimodal slow-thinking methods have demonstrated remarkable success in single-image understanding scenarios, their effectiveness becomes fundamentally constrained when extended to more complex multi-image comprehension tasks. This limitation stems from their predominant reliance on text-based intermediate reasoning processes. While for human, when engaging in sophisticated multi-image analysis, they typically perform two complementary cognitive operations: (1) continuous cross-image visual comparison through region-of-interest matching, and (2) dynamic memorization of critical visual concepts throughout the reasoning chain. Motivated by these observations, we propose the Complex Multi-Modal Chain-of-Thought (CMMCoT) framework, a multi-step reasoning framework that mimics human-like "slow thinking" for multi-image understanding. Our approach incorporates two key innovations: 1. The construction of interleaved multimodal multi-step reasoning chains, which utilize critical visual region tokens, extracted from intermediate reasoning steps, as supervisory signals. This mechanism not only facilitates comprehensive cross-modal understanding but also enhances model interpretability. 2. The introduction of a test-time memory augmentation module that expands the model reasoning capacity during inference while preserving parameter efficiency. Furthermore, to facilitate research in this direction, we have curated a novel multi-image slow-thinking dataset. Extensive experiments demonstrate the effectiveness of our model.

MINT: Multi-modal Chain of Thought in Unified Generative Models for Enhanced Image Generation

Unified generative models have demonstrated extraordinary performance in both text and image generation. However, they tend to underperform when generating intricate images with various interwoven conditions, which is hard to solely rely on straightforward text-to-image generation. In response to this challenge, we introduce MINT, an innovative unified generative model, empowered with native multimodal chain of thought (MCoT) for enhanced image generation for the first time. Firstly, we design Mixture of Transformer Experts (MTXpert), an expert-parallel structure that effectively supports both natural language generation (NLG) and visual capabilities, while avoiding potential modality conflicts that could hinder the full potential of each modality. Building on this, we propose an innovative MCoT training paradigm, a step-by-step approach to multimodal thinking, reasoning, and reflection specifically designed to enhance image generation. This paradigm equips MINT with nuanced, element-wise decoupled alignment and a comprehensive understanding of textual and visual components. Furthermore, it fosters advanced multimodal reasoning and self-reflection, enabling the construction of images that are firmly grounded in the logical relationships between these elements. Notably, MINT has been validated to exhibit superior performance across multiple benchmarks for text-to-image (T2I) and image-to-text (I2T) tasks.

T2I-FactualBench: Benchmarking the Factuality of Text-to-Image Models with Knowledge-Intensive Concepts

Evaluating the quality of synthesized images remains a significant challenge in the development of text-to-image (T2I) generation. Most existing studies in this area primarily focus on evaluating text-image alignment, image quality, and object composition capabilities, with comparatively fewer studies addressing the evaluation of the factuality of T2I models, particularly when the concepts involved are knowledge-intensive. To mitigate this gap, we present T2I-FactualBench in this work - the largest benchmark to date in terms of the number of concepts and prompts specifically designed to evaluate the factuality of knowledge-intensive concept generation. T2I-FactualBench consists of a three-tiered knowledge-intensive text-to-image generation framework, ranging from the basic memorization of individual knowledge concepts to the more complex composition of multiple knowledge concepts. We further introduce a multi-round visual question answering (VQA) based evaluation framework to assess the factuality of three-tiered knowledge-intensive text-to-image generation tasks. Experiments on T2I-FactualBench indicate that current state-of-the-art (SOTA) T2I models still leave significant room for improvement.

SAG: Style-Aligned Article Generation via Model Collaboration

Large language models (LLMs) have increased the demand for personalized and stylish content generation. However, closed-source models like GPT-4 present limitations in optimization opportunities, while the substantial training costs and inflexibility of open-source alternatives, such as Qwen-72B, pose considerable challenges. Conversely, small language models (SLMs) struggle with understanding complex instructions and transferring learned capabilities to new contexts, often exhibiting more pronounced limitations. In this paper, we present a novel collaborative training framework that leverages the strengths of both LLMs and SLMs for style article generation, surpassing the performance of either model alone. We freeze the LLMs to harness their robust instruction-following capabilities and subsequently apply supervised fine-tuning on the SLM using style-specific data. Additionally, we introduce a self-improvement method to enhance style consistency. Our new benchmark, NoteBench, thoroughly evaluates style-aligned generation. Extensive experiments show that our approach achieves state-of-the-art performance, with improvements of 0.78 in ROUGE-L and 0.55 in BLEU-4 scores compared to GPT-4, while maintaining a low hallucination rate regarding factual and faithfulness.

LLaVA-MoD: Making LLaVA Tiny via MoE Knowledge Distillation

We introduce LLaVA-MoD, a novel framework designed to enable the efficient training of small-scale Multimodal Language Models (s-MLLM) by distilling knowledge from large-scale MLLM (l-MLLM). Our approach tackles two fundamental challenges in MLLM distillation. First, we optimize the network structure of s-MLLM by integrating a sparse Mixture of Experts (MoE) architecture into the language model, striking a balance between computational efficiency and model expressiveness. Second, we propose a progressive knowledge transfer strategy to ensure comprehensive knowledge migration. This strategy begins with mimic distillation, where we minimize the Kullback-Leibler (KL) divergence between output distributions to enable the student model to emulate the teacher network's understanding. Following this, we introduce preference distillation via Direct Preference Optimization (DPO), where the key lies in treating l-MLLM as the reference model. During this phase, the s-MLLM's ability to discriminate between superior and inferior examples is significantly enhanced beyond l-MLLM, leading to a better student that surpasses its teacher, particularly in hallucination benchmarks. Extensive experiments demonstrate that LLaVA-MoD outperforms existing models across various multimodal benchmarks while maintaining a minimal number of activated parameters and low computational costs. Remarkably, LLaVA-MoD, with only 2B activated parameters, surpasses Qwen-VL-Chat-7B by an average of 8.8% across benchmarks, using merely 0.3% of the training data and 23% trainable parameters. These results underscore LLaVA-MoD's ability to effectively distill comprehensive knowledge from its teacher model, paving the way for the development of more efficient MLLMs. The code will be available on: https://github.com/shufangxun/LLaVA-MoD.

MARS: Mixture of Auto-Regressive Models for Fine-grained Text-to-image Synthesis

Auto-regressive models have made significant progress in the realm of language generation, yet they do not perform on par with diffusion models in the domain of image synthesis. In this work, we introduce MARS, a novel framework for T2I generation that incorporates a specially designed Semantic Vision-Language Integration Expert (SemVIE). This innovative component integrates pre-trained LLMs by independently processing linguistic and visual information, freezing the textual component while fine-tuning the visual component. This methodology preserves the NLP capabilities of LLMs while imbuing them with exceptional visual understanding. Building upon the powerful base of the pre-trained Qwen-7B, MARS stands out with its bilingual generative capabilities corresponding to both English and Chinese language prompts and the capacity for joint image and text generation. The flexibility of this framework lends itself to migration towards any-to-any task adaptability. Furthermore, MARS employs a multi-stage training strategy that first establishes robust image-text alignment through complementary bidirectional tasks and subsequently concentrates on refining the T2I generation process, significantly augmenting text-image synchrony and the granularity of image details. Notably, MARS requires only 9% of the GPU days needed by SD1.5, yet it achieves remarkable results across a variety of benchmarks, illustrating the training efficiency and the potential for swift deployment in various applications.

Autoregressive Pretraining with Mamba in Vision

The vision community has started to build with the recently developed state space model, Mamba, as the new backbone for a range of tasks. This paper shows that Mamba's visual capability can be significantly enhanced through autoregressive pretraining, a direction not previously explored. Efficiency-wise, the autoregressive nature can well capitalize on the Mamba's unidirectional recurrent structure, enabling faster overall training speed compared to other training strategies like mask modeling. Performance-wise, autoregressive pretraining equips the Mamba architecture with markedly higher accuracy over its supervised-trained counterparts and, more importantly, successfully unlocks its scaling potential to large and even huge model sizes. For example, with autoregressive pretraining, a base-size Mamba attains 83.2\% ImageNet accuracy, outperforming its supervised counterpart by 2.0\%; our huge-size Mamba, the largest Vision Mamba to date, attains 85.0\% ImageNet accuracy (85.5\% when finetuned with $384\times384$ inputs), notably surpassing all other Mamba variants in vision. The code is available at \url{https://github.com/OliverRensu/ARM}.

HyperLLaVA: Dynamic Visual and Language Expert Tuning for Multimodal Large Language Models

Recent advancements indicate that scaling up Multimodal Large Language Models (MLLMs) effectively enhances performance on downstream multimodal tasks. The prevailing MLLM paradigm, \emph{e.g.}, LLaVA, transforms visual features into text-like tokens using a \emph{static} vision-language mapper, thereby enabling \emph{static} LLMs to develop the capability to comprehend visual information through visual instruction tuning. Although promising, the \emph{static} tuning strategy~\footnote{The static tuning refers to the trained model with static parameters.} that shares the same parameters may constrain performance across different downstream multimodal tasks. In light of this, we introduce HyperLLaVA, which involves adaptive tuning of the projector and LLM parameters, in conjunction with a dynamic visual expert and language expert, respectively. These experts are derived from HyperNetworks, which generates adaptive parameter shifts through visual and language guidance, enabling dynamic projector and LLM modeling in two-stage training. Our experiments demonstrate that our solution significantly surpasses LLaVA on existing MLLM benchmarks, including MME, MMBench, SEED-Bench, and LLaVA-Bench. ~\footnote{Our project is available on the link https://github.com/DCDmllm/HyperLLaVA}.

Audio-Visual LLM for Video Understanding

This paper presents Audio-Visual LLM, a Multimodal Large Language Model that takes both visual and auditory inputs for holistic video understanding. A key design is the modality-augmented training, which involves the integration of modality-specific tokens engineered to activate the appropriate visual and/or auditory encoder selectively. This mechanism is pivotal in enabling end-to-end joint training with video data at different modalities, including visual-only, audio-only, and audio-visual formats. Moreover, we introduce a high-quality video instruction dataset, derived from GPT-4. This dataset allows Audio-Visual LLM to adeptly process a variety of task-oriented video instructions, ranging from multi-turn conversations and audio-visual narratives to complex reasoning tasks. Extensive experiments demonstrate that Audio-Visual LLM impressively achieves strong zero-shot results across a range of video understanding tasks. For example, Audio-Visual LLM achieves an accuracy of 53.7% on MSRVTT-QA, outperforming non-LLM-based InterVideo by 6.6% and LLM-based Valley by 4.4%, respectively. Additionally, our Audio-Visual LLM also achieves competitive performance on audio tasks (e.g., AudioCaps).

Compress & Align: Curating Image-Text Data with Human Knowledge

The massive growth of image-text data through web crawling inherently presents the challenge of variability in data quality. This paper introduces a novel algorithm, rooted in human knowledge, to compress this vast corpus of web-crawled image-text datasets to a compact and high-quality form. Our method unfolds in three major steps. First, we collect an image-text dataset, wherein each image is associated with multiple captions sourced from diverse origins. Then, to systemically capture human preferences regarding the best caption paired with each image, we establish a comprehensive set of both subjective and objective criteria for critically guiding the alignment assessment from labelers. Lastly, we train a reward model on the annotated dataset to internalize the nuanced human understanding of image-text alignment. The resulting reward model thus can act as a human-like referee to filter misaligned/low-quality image-text pairs. Extensive experiments demonstrate that we are able to secure (or even improve) model performance by compressing the image-text datasets up to ~90%. An impressive example is that, by aggressively reducing the total training sample from 130M to 15.5M (e.g., ~9x smaller), our BLIP-B/16 models still consistently show superior performance compared with the full-size-dataset counterpart on image-text retrieval (Flickr30K, COCO) by ~2.5% in Recall@1, and on image-captioning (Nocaps, COCO) by ~10.0% in CIDEr and ~2.7% in SPICE.