Renrui Zhang

Renrui Zhang, Jiaming Han, Chris Liu et al. · stanford

We present LLaMA-Adapter, a lightweight adaption method to efficiently fine-tune LLaMA into an instruction-following model. Using 52K self-instruct demonstrations, LLaMA-Adapter only introduces 1.2M learnable parameters upon the frozen LLaMA 7B model, and costs less than one hour for fine-tuning on 8 A100 GPUs. Specifically, we adopt a set of learnable adaption prompts, and prepend them to the word tokens at higher transformer layers. Then, a zero-initialized attention mechanism with zero gating is proposed, which adaptively injects the new instructional cues into LLaMA, while effectively preserves its pre-trained knowledge. With our efficient training, LLaMA-Adapter can generate high-quality responses, comparable to Alpaca with fully fine-tuned 7B parameters. Besides language commands, our approach can be simply extended to multi-modal instructions for learning image-conditioned LLaMA model, which achieves superior reasoning performance on ScienceQA and COCO Caption benchmarks. Furthermore, we also evaluate the zero-initialized attention mechanism for fine-tuning other pre-trained models (ViT, RoBERTa) on traditional vision and language tasks, demonstrating the superior generalization capacity of our approach. Code is released at https://github.com/OpenGVLab/LLaMA-Adapter.

Peng Gao, Jiaming Han, Renrui Zhang et al. · berkeley, stanford

How to efficiently transform large language models (LLMs) into instruction followers is recently a popular research direction, while training LLM for multi-modal reasoning remains less explored. Although the recent LLaMA-Adapter demonstrates the potential to handle visual inputs with LLMs, it still cannot generalize well to open-ended visual instructions and lags behind GPT-4. In this paper, we present LLaMA-Adapter V2, a parameter-efficient visual instruction model. Specifically, we first augment LLaMA-Adapter by unlocking more learnable parameters (e.g., norm, bias and scale), which distribute the instruction-following ability across the entire LLaMA model besides adapters. Secondly, we propose an early fusion strategy to feed visual tokens only into the early LLM layers, contributing to better visual knowledge incorporation. Thirdly, a joint training paradigm of image-text pairs and instruction-following data is introduced by optimizing disjoint groups of learnable parameters. This strategy effectively alleviates the interference between the two tasks of image-text alignment and instruction following and achieves strong multi-modal reasoning with only a small-scale image-text and instruction dataset. During inference, we incorporate additional expert models (e.g. captioning/OCR systems) into LLaMA-Adapter to further enhance its image understanding capability without incurring training costs. Compared to the original LLaMA-Adapter, our LLaMA-Adapter V2 can perform open-ended multi-modal instructions by merely introducing 14M parameters over LLaMA. The newly designed framework also exhibits stronger language-only instruction-following capabilities and even excels in chat interactions. Our code and models are available at https://github.com/ZrrSkywalker/LLaMA-Adapter.

Jiaming Han, Renrui Zhang, Wenqi Shao et al. · berkeley

We present ImageBind-LLM, a multi-modality instruction tuning method of large language models (LLMs) via ImageBind. Existing works mainly focus on language and image instruction tuning, different from which, our ImageBind-LLM can respond to multi-modality conditions, including audio, 3D point clouds, video, and their embedding-space arithmetic by only image-text alignment training. During training, we adopt a learnable bind network to align the embedding space between LLaMA and ImageBind's image encoder. Then, the image features transformed by the bind network are added to word tokens of all layers in LLaMA, which progressively injects visual instructions via an attention-free and zero-initialized gating mechanism. Aided by the joint embedding of ImageBind, the simple image-text training enables our model to exhibit superior multi-modality instruction-following capabilities. During inference, the multi-modality inputs are fed into the corresponding ImageBind encoders, and processed by a proposed visual cache model for further cross-modal embedding enhancement. The training-free cache model retrieves from three million image features extracted by ImageBind, which effectively mitigates the training-inference modality discrepancy. Notably, with our approach, ImageBind-LLM can respond to instructions of diverse modalities and demonstrate significant language generation quality. Code is released at https://github.com/OpenGVLab/LLaMA-Adapter.

Renrui Zhang, Zhang Wei, Rongyao Fang et al.

Contrastive Vision-Language Pre-training, known as CLIP, has provided a new paradigm for learning visual representations using large-scale image-text pairs. It shows impressive performance on downstream tasks by zero-shot knowledge transfer. To further enhance CLIP's adaption capability, existing methods proposed to fine-tune additional learnable modules, which significantly improves the few-shot performance but introduces extra training time and computational resources. In this paper, we propose a training-free adaption method for CLIP to conduct few-shot classification, termed as Tip-Adapter, which not only inherits the training-free advantage of zero-shot CLIP but also performs comparably to those training-required approaches. Tip-Adapter constructs the adapter via a key-value cache model from the few-shot training set, and updates the prior knowledge encoded in CLIP by feature retrieval. On top of that, the performance of Tip-Adapter can be further boosted to be state-of-the-art on ImageNet by fine-tuning the cache model for 10$\times$ fewer epochs than existing methods, which is both effective and efficient. We conduct extensive experiments of few-shot classification on 11 datasets to demonstrate the superiority of our proposed methods. Code is released at https://github.com/gaopengcuhk/Tip-Adapter.

Renrui Zhang, Ziyu Guo, Rongyao Fang et al.

Masked Autoencoders (MAE) have shown great potentials in self-supervised pre-training for language and 2D image transformers. However, it still remains an open question on how to exploit masked autoencoding for learning 3D representations of irregular point clouds. In this paper, we propose Point-M2AE, a strong Multi-scale MAE pre-training framework for hierarchical self-supervised learning of 3D point clouds. Unlike the standard transformer in MAE, we modify the encoder and decoder into pyramid architectures to progressively model spatial geometries and capture both fine-grained and high-level semantics of 3D shapes. For the encoder that downsamples point tokens by stages, we design a multi-scale masking strategy to generate consistent visible regions across scales, and adopt a local spatial self-attention mechanism during fine-tuning to focus on neighboring patterns. By multi-scale token propagation, the lightweight decoder gradually upsamples point tokens with complementary skip connections from the encoder, which further promotes the reconstruction from a global-to-local perspective. Extensive experiments demonstrate the state-of-the-art performance of Point-M2AE for 3D representation learning. With a frozen encoder after pre-training, Point-M2AE achieves 92.9% accuracy for linear SVM on ModelNet40, even surpassing some fully trained methods. By fine-tuning on downstream tasks, Point-M2AE achieves 86.43% accuracy on ScanObjectNN, +3.36% to the second-best, and largely benefits the few-shot classification, part segmentation and 3D object detection with the hierarchical pre-training scheme. Code is available at https://github.com/ZrrSkywalker/Point-M2AE.

Yanmin Wu, Xinhua Cheng, Renrui Zhang et al. · pku

3D visual grounding aims to find the object within point clouds mentioned by free-form natural language descriptions with rich semantic cues. However, existing methods either extract the sentence-level features coupling all words or focus more on object names, which would lose the word-level information or neglect other attributes. To alleviate these issues, we present EDA that Explicitly Decouples the textual attributes in a sentence and conducts Dense Alignment between such fine-grained language and point cloud objects. Specifically, we first propose a text decoupling module to produce textual features for every semantic component. Then, we design two losses to supervise the dense matching between two modalities: position alignment loss and semantic alignment loss. On top of that, we further introduce a new visual grounding task, locating objects without object names, which can thoroughly evaluate the model's dense alignment capacity. Through experiments, we achieve state-of-the-art performance on two widely-adopted 3D visual grounding datasets, ScanRefer and SR3D/NR3D, and obtain absolute leadership on our newly-proposed task. The source code is available at https://github.com/yanmin-wu/EDA.

Renrui Zhang, Xiangfei Hu, Bohao Li et al.

Visual recognition in low-data regimes requires deep neural networks to learn generalized representations from limited training samples. Recently, CLIP-based methods have shown promising few-shot performance benefited from the contrastive language-image pre-training. We then question, if the more diverse pre-training knowledge can be cascaded to further assist few-shot representation learning. In this paper, we propose CaFo, a Cascade of Foundation models that incorporates diverse prior knowledge of various pre-training paradigms for better few-shot learning. Our CaFo incorporates CLIP's language-contrastive knowledge, DINO's vision-contrastive knowledge, DALL-E's vision-generative knowledge, and GPT-3's language-generative knowledge. Specifically, CaFo works by 'Prompt, Generate, then Cache'. Firstly, we leverage GPT-3 to produce textual inputs for prompting CLIP with rich downstream linguistic semantics. Then, we generate synthetic images via DALL-E to expand the few-shot training data without any manpower. At last, we introduce a learnable cache model to adaptively blend the predictions from CLIP and DINO. By such collaboration, CaFo can fully unleash the potential of different pre-training methods and unify them to perform state-of-the-art for few-shot classification. Code is available at https://github.com/ZrrSkywalker/CaFo.

Renrui Zhang, Liuhui Wang, Yu Qiao et al.

Pre-training by numerous image data has become de-facto for robust 2D representations. In contrast, due to the expensive data acquisition and annotation, a paucity of large-scale 3D datasets severely hinders the learning for high-quality 3D features. In this paper, we propose an alternative to obtain superior 3D representations from 2D pre-trained models via Image-to-Point Masked Autoencoders, named as I2P-MAE. By self-supervised pre-training, we leverage the well learned 2D knowledge to guide 3D masked autoencoding, which reconstructs the masked point tokens with an encoder-decoder architecture. Specifically, we first utilize off-the-shelf 2D models to extract the multi-view visual features of the input point cloud, and then conduct two types of image-to-point learning schemes on top. For one, we introduce a 2D-guided masking strategy that maintains semantically important point tokens to be visible for the encoder. Compared to random masking, the network can better concentrate on significant 3D structures and recover the masked tokens from key spatial cues. For another, we enforce these visible tokens to reconstruct the corresponding multi-view 2D features after the decoder. This enables the network to effectively inherit high-level 2D semantics learned from rich image data for discriminative 3D modeling. Aided by our image-to-point pre-training, the frozen I2P-MAE, without any fine-tuning, achieves 93.4% accuracy for linear SVM on ModelNet40, competitive to the fully trained results of existing methods. By further fine-tuning on on ScanObjectNN's hardest split, I2P-MAE attains the state-of-the-art 90.11% accuracy, +3.68% to the second-best, demonstrating superior transferable capacity. Code will be available at https://github.com/ZrrSkywalker/I2P-MAE.

Renrui Zhang, Han Qiu, Tai Wang et al.

Monocular 3D object detection has long been a challenging task in autonomous driving. Most existing methods follow conventional 2D detectors to first localize object centers, and then predict 3D attributes by neighboring features. However, only using local visual features is insufficient to understand the scene-level 3D spatial structures and ignores the long-range inter-object depth relations. In this paper, we introduce the first DETR framework for Monocular DEtection with a depth-guided TRansformer, named MonoDETR. We modify the vanilla transformer to be depth-aware and guide the whole detection process by contextual depth cues. Specifically, concurrent to the visual encoder that captures object appearances, we introduce to predict a foreground depth map, and specialize a depth encoder to extract non-local depth embeddings. Then, we formulate 3D object candidates as learnable queries and propose a depth-guided decoder to conduct object-scene depth interactions. In this way, each object query estimates its 3D attributes adaptively from the depth-guided regions on the image and is no longer constrained to local visual features. On KITTI benchmark with monocular images as input, MonoDETR achieves state-of-the-art performance and requires no extra dense depth annotations. Besides, our depth-guided modules can also be plug-and-play to enhance multi-view 3D object detectors on nuScenes dataset, demonstrating our superior generalization capacity. Code is available at https://github.com/ZrrSkywalker/MonoDETR.

Ziyi Lin, Shijie Geng, Renrui Zhang et al.

Video recognition has been dominated by the end-to-end learning paradigm -- first initializing a video recognition model with weights of a pretrained image model and then conducting end-to-end training on videos. This enables the video network to benefit from the pretrained image model. However, this requires substantial computation and memory resources for finetuning on videos and the alternative of directly using pretrained image features without finetuning the image backbone leads to subpar results. Fortunately, recent advances in Contrastive Vision-Language Pre-training (CLIP) pave the way for a new route for visual recognition tasks. Pretrained on large open-vocabulary image-text pair data, these models learn powerful visual representations with rich semantics. In this paper, we present Efficient Video Learning (EVL) -- an efficient framework for directly training high-quality video recognition models with frozen CLIP features. Specifically, we employ a lightweight Transformer decoder and learn a query token to dynamically collect frame-level spatial features from the CLIP image encoder. Furthermore, we adopt a local temporal module in each decoder layer to discover temporal clues from adjacent frames and their attention maps. We show that despite being efficient to train with a frozen backbone, our models learn high quality video representations on a variety of video recognition datasets. Code is available at https://github.com/OpenGVLab/efficient-video-recognition.

Renrui Zhang, Liuhui Wang, Ziyu Guo et al.

We present a Non-parametric Network for 3D point cloud analysis, Point-NN, which consists of purely non-learnable components: farthest point sampling (FPS), k-nearest neighbors (k-NN), and pooling operations, with trigonometric functions. Surprisingly, it performs well on various 3D tasks, requiring no parameters or training, and even surpasses existing fully trained models. Starting from this basic non-parametric model, we propose two extensions. First, Point-NN can serve as a base architectural framework to construct Parametric Networks by simply inserting linear layers on top. Given the superior non-parametric foundation, the derived Point-PN exhibits a high performance-efficiency trade-off with only a few learnable parameters. Second, Point-NN can be regarded as a plug-and-play module for the already trained 3D models during inference. Point-NN captures the complementary geometric knowledge and enhances existing methods for different 3D benchmarks without re-training. We hope our work may cast a light on the community for understanding 3D point clouds with non-parametric methods. Code is available at https://github.com/ZrrSkywalker/Point-NN.

Dongzhi Jiang, Renrui Zhang, Ziyu Guo et al. · pku

The advent of Large Language Models (LLMs) has paved the way for AI search engines, e.g., SearchGPT, showcasing a new paradigm in human-internet interaction. However, most current AI search engines are limited to text-only settings, neglecting the multimodal user queries and the text-image interleaved nature of website information. Recently, Large Multimodal Models (LMMs) have made impressive strides. Yet, whether they can function as AI search engines remains under-explored, leaving the potential of LMMs in multimodal search an open question. To this end, we first design a delicate pipeline, MMSearch-Engine, to empower any LMMs with multimodal search capabilities. On top of this, we introduce MMSearch, a comprehensive evaluation benchmark to assess the multimodal search performance of LMMs. The curated dataset contains 300 manually collected instances spanning 14 subfields, which involves no overlap with the current LMMs' training data, ensuring the correct answer can only be obtained within searching. By using MMSearch-Engine, the LMMs are evaluated by performing three individual tasks (requery, rerank, and summarization), and one challenging end-to-end task with a complete searching process. We conduct extensive experiments on closed-source and open-source LMMs. Among all tested models, GPT-4o with MMSearch-Engine achieves the best results, which surpasses the commercial product, Perplexity Pro, in the end-to-end task, demonstrating the effectiveness of our proposed pipeline. We further present error analysis to unveil current LMMs still struggle to fully grasp the multimodal search tasks, and conduct ablation study to indicate the potential of scaling test-time computation for AI search engine. We hope MMSearch may provide unique insights to guide the future development of multimodal AI search engine. Project Page: https://mmsearch.github.io

Ziyi Lin, Chris Liu, Renrui Zhang et al.

We present SPHINX, a versatile multi-modal large language model (MLLM) with a joint mixing of model weights, tuning tasks, and visual embeddings. First, for stronger vision-language alignment, we unfreeze the large language model (LLM) during pre-training, and introduce a weight mix strategy between LLMs trained by real-world and synthetic data. By directly integrating the weights from two domains, the mixed LLM can efficiently incorporate diverse semantics with favorable robustness. Then, to enable multi-purpose capabilities, we mix a variety of tasks for joint visual instruction tuning, and design task-specific instructions to avoid inter-task conflict. In addition to the basic visual question answering, we include more challenging tasks such as region-level understanding, caption grounding, document layout detection, and human pose estimation, contributing to mutual enhancement over different scenarios. Additionally, we propose to extract comprehensive visual embeddings from various network architectures, pre-training paradigms, and information granularity, providing language models with more robust image representations. Based on our proposed joint mixing, SPHINX exhibits superior multi-modal understanding capabilities on a wide range of applications. On top of this, we further propose an efficient strategy aiming to better capture fine-grained appearances of high-resolution images. With a mixing of different scales and high-resolution sub-images, SPHINX attains exceptional visual parsing and reasoning performance on existing evaluation benchmarks. We hope our work may cast a light on the exploration of joint mixing in future MLLM research. Code is released at https://github.com/Alpha-VLLM/LLaMA2-Accessory.

Peng Gao, Renrui Zhang, Rongyao Fang et al.

Masked Autoencoders (MAE) have been popular paradigms for large-scale vision representation pre-training. However, MAE solely reconstructs the low-level RGB signals after the decoder and lacks supervision upon high-level semantics for the encoder, thus suffering from sub-optimal learned representations and long pre-training epochs. To alleviate this, previous methods simply replace the pixel reconstruction targets of 75% masked tokens by encoded features from pre-trained image-image (DINO) or image-language (CLIP) contrastive learning. Different from those efforts, we propose to Mimic before Reconstruct for Masked Autoencoders, named as MR-MAE, which jointly learns high-level and low-level representations without interference during pre-training. For high-level semantics, MR-MAE employs a mimic loss over 25% visible tokens from the encoder to capture the pre-trained patterns encoded in CLIP and DINO. For low-level structures, we inherit the reconstruction loss in MAE to predict RGB pixel values for 75% masked tokens after the decoder. As MR-MAE applies high-level and low-level targets respectively at different partitions, the learning conflicts between them can be naturally overcome and contribute to superior visual representations for various downstream tasks. On ImageNet-1K, the MR-MAE base pre-trained for only 400 epochs achieves 85.8% top-1 accuracy after fine-tuning, surpassing the 1600-epoch MAE base by +2.2% and the previous state-of-the-art BEiT V2 base by +0.3%. Code and pre-trained models will be released at https://github.com/Alpha-VL/ConvMAE.

Keqiang Sun, Junting Pan, Yuying Ge et al. · pku

While recent advancements in vision-language models have had a transformative impact on multi-modal comprehension, the extent to which these models possess the ability to comprehend generated images remains uncertain. Synthetic images, in comparison to real data, encompass a higher level of diversity in terms of both content and style, thereby presenting significant challenges for the models to fully grasp. In light of this challenge, we introduce a comprehensive dataset, referred to as JourneyDB, that caters to the domain of generative images within the context of multi-modal visual understanding. Our meticulously curated dataset comprises 4 million distinct and high-quality generated images, each paired with the corresponding text prompts that were employed in their creation. Furthermore, we additionally introduce an external subset with results of another 22 text-to-image generative models, which makes JourneyDB a comprehensive benchmark for evaluating the comprehension of generated images. On our dataset, we have devised four benchmarks to assess the performance of generated image comprehension in relation to both content and style interpretation. These benchmarks encompass prompt inversion, style retrieval, image captioning, and visual question answering. Lastly, we evaluate the performance of state-of-the-art multi-modal models when applied to the JourneyDB dataset, providing a comprehensive analysis of their strengths and limitations in comprehending generated content. We anticipate that the proposed dataset and benchmarks will facilitate further research in the field of generative content understanding. The dataset is publicly available at https://journeydb.github.io.

Ke Wang, Houxing Ren, Aojun Zhou et al.

The recently released GPT-4 Code Interpreter has demonstrated remarkable proficiency in solving challenging math problems, primarily attributed to its ability to seamlessly reason with natural language, generate code, execute code, and continue reasoning based on the execution output. In this paper, we present a method to fine-tune open-source language models, enabling them to use code for modeling and deriving math equations and, consequently, enhancing their mathematical reasoning abilities. We propose a method of generating novel and high-quality datasets with math problems and their code-based solutions, referred to as MathCodeInstruct. Each solution interleaves natural language, code, and execution results. We also introduce a customized supervised fine-tuning and inference approach. This approach yields the MathCoder models, a family of models capable of generating code-based solutions for solving challenging math problems. Impressively, the MathCoder models achieve state-of-the-art scores among open-source LLMs on the MATH (45.2%) and GSM8K (83.9%) datasets, substantially outperforming other open-source alternatives. Notably, the MathCoder model not only surpasses ChatGPT-3.5 and PaLM-2 on GSM8K and MATH but also outperforms GPT-4 on the competition-level MATH dataset. The dataset and models will be released at https://github.com/mathllm/MathCoder.

Feng Li, Renrui Zhang, Hao Zhang et al.

Visual instruction tuning has made considerable strides in enhancing the capabilities of Large Multimodal Models (LMMs). However, existing open LMMs largely focus on single-image tasks, their applications to multi-image scenarios remains less explored. Additionally, prior LMM research separately tackles different scenarios, leaving it impossible to generalize cross scenarios with new emerging capabilities. To this end, we introduce LLaVA-NeXT-Interleave, which simultaneously tackles Multi-image, Multi-frame (video), Multi-view (3D), and Multi-patch (single-image) scenarios in LMMs. To enable these capabilities, we regard the interleaved data format as a general template and compile the M4-Instruct dataset with 1,177.6k samples, spanning 4 primary domains with 14 tasks and 41 datasets. We also curate the LLaVA-Interleave Bench to comprehensively evaluate the multi-image performance of LMMs. Through extensive experiments, LLaVA-NeXT-Interleave achieves leading results in multi-image, video, and 3D benchmarks, while maintaining the performance of single-image tasks. Besides, our model also exhibits several emerging capabilities, e.g., transferring tasks across different settings and modalities. Code is available at https://github.com/LLaVA-VL/LLaVA-NeXT

Anthony Chen, Kevin Zhang, Renrui Zhang et al.

Masked Autoencoders learn strong visual representations and achieve state-of-the-art results in several independent modalities, yet very few works have addressed their capabilities in multi-modality settings. In this work, we focus on point cloud and RGB image data, two modalities that are often presented together in the real world, and explore their meaningful interactions. To improve upon the cross-modal synergy in existing works, we propose PiMAE, a self-supervised pre-training framework that promotes 3D and 2D interaction through three aspects. Specifically, we first notice the importance of masking strategies between the two sources and utilize a projection module to complementarily align the mask and visible tokens of the two modalities. Then, we utilize a well-crafted two-branch MAE pipeline with a novel shared decoder to promote cross-modality interaction in the mask tokens. Finally, we design a unique cross-modal reconstruction module to enhance representation learning for both modalities. Through extensive experiments performed on large-scale RGB-D scene understanding benchmarks (SUN RGB-D and ScannetV2), we discover it is nontrivial to interactively learn point-image features, where we greatly improve multiple 3D detectors, 2D detectors, and few-shot classifiers by 2.9%, 6.7%, and 2.4%, respectively. Code is available at https://github.com/BLVLab/PiMAE.

Renrui Zhang, Ziyao Zeng, Ziyu Guo et al.

Besides image classification, Contrastive Language-Image Pre-training (CLIP) has accomplished extraordinary success for a wide range of vision tasks, including object-level and 3D space understanding. However, it's still challenging to transfer semantic knowledge learned from CLIP into more intricate tasks of quantified targets, such as depth estimation with geometric information. In this paper, we propose to apply CLIP for zero-shot monocular depth estimation, named DepthCLIP. We found that the patches of the input image could respond to a certain semantic distance token and then be projected to a quantified depth bin for coarse estimation. Without any training, our DepthCLIP surpasses existing unsupervised methods and even approaches the early fully-supervised networks. To our best knowledge, we are the first to conduct zero-shot adaptation from the semantic language knowledge to quantified downstream tasks and perform zero-shot monocular depth estimation. We hope our work could cast a light on future research. The code is available at https://github.com/Adonis-galaxy/DepthCLIP.

Peixiang Huang, Li Liu, Renrui Zhang et al.

To achieve accurate and low-cost 3D object detection, existing methods propose to benefit camera-based multi-view detectors with spatial cues provided by the LiDAR modality, e.g., dense depth supervision and bird-eye-view (BEV) feature distillation. However, they directly conduct point-to-point mimicking from LiDAR to camera, which neglects the inner-geometry of foreground targets and suffers from the modal gap between 2D-3D features. In this paper, we propose the learning scheme of Target Inner-Geometry from the LiDAR modality into camera-based BEV detectors for both dense depth and BEV features, termed as TiG-BEV. First, we introduce an inner-depth supervision module to learn the low-level relative depth relations between different foreground pixels. This enables the camera-based detector to better understand the object-wise spatial structures. Second, we design an inner-feature BEV distillation module to imitate the high-level semantics of different keypoints within foreground targets. To further alleviate the BEV feature gap between two modalities, we adopt both inter-channel and inter-keypoint distillation for feature-similarity modeling. With our target inner-geometry distillation, TiG-BEV can effectively boost BEVDepth by +2.3% NDS and +2.4% mAP, along with BEVDet by +9.1% NDS and +10.3% mAP on nuScenes val set. Code will be available at https://github.com/ADLab3Ds/TiG-BEV.

Kexue Fu, Peng Gao, ShaoLei Liu et al.

Recently, the pre-training paradigm combining Transformer and masked language modeling has achieved tremendous success in NLP, images, and point clouds, such as BERT. However, directly extending BERT from NLP to point clouds requires training a fixed discrete Variational AutoEncoder (dVAE) before pre-training, which results in a complex two-stage method called Point-BERT. Inspired by BERT and MoCo, we propose POS-BERT, a one-stage BERT pre-training method for point clouds. Specifically, we use the mask patch modeling (MPM) task to perform point cloud pre-training, which aims to recover masked patches information under the supervision of the corresponding tokenizer output. Unlike Point-BERT, its tokenizer is extra-trained and frozen. We propose to use the dynamically updated momentum encoder as the tokenizer, which is updated and outputs the dynamic supervision signal along with the training process. Further, in order to learn high-level semantic representation, we combine contrastive learning to maximize the class token consistency between different transformation point clouds. Extensive experiments have demonstrated that POS-BERT can extract high-quality pre-training features and promote downstream tasks to improve performance. Using the pre-training model without any fine-tuning to extract features and train linear SVM on ModelNet40, POS-BERT achieves the state-of-the-art classification accuracy, which exceeds Point-BERT by 3.5\%. In addition, our approach has significantly improved many downstream tasks, such as fine-tuned classification, few-shot classification, part segmentation. The code and trained-models will be available at: \url{https://github.com/fukexue/POS-BERT}.

Ziteng Cui, Yingying Zhu, Lin Gu et al.

Image restoration algorithms such as super resolution (SR) are indispensable pre-processing modules for object detection in low quality images. Most of these algorithms assume the degradation is fixed and known a priori. However, in practical, either the real degradation or optimal up-sampling ratio rate is unknown or differs from assumption, leading to a deteriorating performance for both the pre-processing module and the consequent high-level task such as object detection. Here, we propose a novel self-supervised framework to detect objects in degraded low resolution images. We utilizes the downsampling degradation as a kind of transformation for self-supervised signals to explore the equivariant representation against various resolutions and other degradation conditions. The Auto Encoding Resolution in Self-supervision (AERIS) framework could further take the advantage of advanced SR architectures with an arbitrary resolution restoring decoder to reconstruct the original correspondence from the degraded input image. Both the representation learning and object detection are optimized jointly in an end-to-end training fashion. The generic AERIS framework could be implemented on various mainstream object detection architectures with different backbones. The extensive experiments show that our methods has achieved superior performance compared with existing methods when facing variant degradation situations. Code would be released at https://github.com/cuiziteng/ECCV_AERIS.

Yulu Gan, Yan Bai, Yihang Lou et al. · pku

Continual Test-Time Adaptation (CTTA) aims to adapt the source model to continually changing unlabeled target domains without access to the source data. Existing methods mainly focus on model-based adaptation in a self-training manner, such as predicting pseudo labels for new domain datasets. Since pseudo labels are noisy and unreliable, these methods suffer from catastrophic forgetting and error accumulation when dealing with dynamic data distributions. Motivated by the prompt learning in NLP, in this paper, we propose to learn an image-level visual domain prompt for target domains while having the source model parameters frozen. During testing, the changing target datasets can be adapted to the source model by reformulating the input data with the learned visual prompts. Specifically, we devise two types of prompts, i.e., domains-specific prompts and domains-agnostic prompts, to extract current domain knowledge and maintain the domain-shared knowledge in the continual adaptation. Furthermore, we design a homeostasis-based prompt adaptation strategy to suppress domain-sensitive parameters in domain-invariant prompts to learn domain-shared knowledge more effectively. This transition from the model-dependent paradigm to the model-free one enables us to bypass the catastrophic forgetting and error accumulation problems. Experiments show that our proposed method achieves significant performance gains over state-of-the-art methods on four widely-used benchmarks, including CIFAR-10C, CIFAR-100C, ImageNet-C, and VLCS datasets.

Jiaben Chen, Renrui Zhang, Dongze Lian et al. · tsinghua

Current audio-visual separation methods share a standard architecture design where an audio encoder-decoder network is fused with visual encoding features at the encoder bottleneck. This design confounds the learning of multi-modal feature encoding with robust sound decoding for audio separation. To generalize to a new instrument: one must finetune the entire visual and audio network for all musical instruments. We re-formulate visual-sound separation task and propose Instrument as Query (iQuery) with a flexible query expansion mechanism. Our approach ensures cross-modal consistency and cross-instrument disentanglement. We utilize "visually named" queries to initiate the learning of audio queries and use cross-modal attention to remove potential sound source interference at the estimated waveforms. To generalize to a new instrument or event class, drawing inspiration from the text-prompt design, we insert an additional query as an audio prompt while freezing the attention mechanism. Experimental results on three benchmarks demonstrate that our iQuery improves audio-visual sound source separation performance.

Weifeng Lin, Xinyu Wei, Renrui Zhang et al.

This paper presents a versatile image-to-image visual assistant, PixWizard, designed for image generation, manipulation, and translation based on free-from language instructions. To this end, we tackle a variety of vision tasks into a unified image-text-to-image generation framework and curate an Omni Pixel-to-Pixel Instruction-Tuning Dataset. By constructing detailed instruction templates in natural language, we comprehensively include a large set of diverse vision tasks such as text-to-image generation, image restoration, image grounding, dense image prediction, image editing, controllable generation, inpainting/outpainting, and more. Furthermore, we adopt Diffusion Transformers (DiT) as our foundation model and extend its capabilities with a flexible any resolution mechanism, enabling the model to dynamically process images based on the aspect ratio of the input, closely aligning with human perceptual processes. The model also incorporates structure-aware and semantic-aware guidance to facilitate effective fusion of information from the input image. Our experiments demonstrate that PixWizard not only shows impressive generative and understanding abilities for images with diverse resolutions but also exhibits promising generalization capabilities with unseen tasks and human instructions. The code and related resources are available at https://github.com/AFeng-x/PixWizard

Renrui Zhang, Xinyu Wei, Dongzhi Jiang et al.

The mathematical capabilities of Multi-modal Large Language Models (MLLMs) remain under-explored with three areas to be improved: visual encoding of math diagrams, diagram-language alignment, and chain-of-thought (CoT) reasoning. This draws forth an urgent demand for an effective training paradigm and a large-scale, comprehensive dataset with detailed CoT rationales, which is challenging to collect and costly to annotate manually. To tackle this issue, we propose MAVIS, a MAthematical VISual instruction tuning pipeline for MLLMs, featuring an automatic data engine to efficiently create mathematical visual datasets. We design the data generation process to be entirely independent of human intervention or GPT API usage, while ensuring the diagram-caption correspondence, question-answer correctness, and CoT reasoning quality. With this approach, we curate two datasets, MAVIS-Caption (558K diagram-caption pairs) and MAVIS-Instruct (834K visual math problems with CoT rationales), and propose four progressive stages for training MLLMs from scratch. First, we utilize MAVIS-Caption to fine-tune a math-specific vision encoder (CLIP-Math) through contrastive learning, tailored for improved diagram visual encoding. Second, we also leverage MAVIS-Caption to align the CLIP-Math with a large language model (LLM) by a projection layer, enhancing vision-language alignment in mathematical domains. Third, we adopt MAVIS-Instruct to perform the instruction tuning for robust problem-solving skills, and term the resulting model as MAVIS-7B. Fourth, we apply Direct Preference Optimization (DPO) to enhance the CoT capabilities of our model, further refining its step-wise reasoning performance. Code and data will be released at https://github.com/ZrrSkywalker/MAVIS

Xiaobao Wei, Renrui Zhang, Jiarui Wu et al.

Neural 3D reconstruction from multi-view images has recently attracted increasing attention from the community. Existing methods normally learn a neural field for the whole scene, while it is still under-explored how to reconstruct a target object indicated by users. Considering the Segment Anything Model (SAM) has shown effectiveness in segmenting any 2D images, in this paper, we propose NTO3D, a novel high-quality Neural Target Object 3D (NTO3D) reconstruction method, which leverages the benefits of both neural field and SAM. We first propose a novel strategy to lift the multi-view 2D segmentation masks of SAM into a unified 3D occupancy field. The 3D occupancy field is then projected into 2D space and generates the new prompts for SAM. This process is iterative until convergence to separate the target object from the scene. After this, we then lift the 2D features of the SAM encoder into a 3D feature field in order to improve the reconstruction quality of the target object. NTO3D lifts the 2D masks and features of SAM into the 3D neural field for high-quality neural target object 3D reconstruction. We conduct detailed experiments on several benchmark datasets to demonstrate the advantages of our method. The code will be available at: https://github.com/ucwxb/NTO3D.

Ziyu Guo, Renrui Zhang, Longtian Qiu et al.

Masked Autoencoders (MAE) have shown promising performance in self-supervised learning for both 2D and 3D computer vision. However, existing MAE-style methods can only learn from the data of a single modality, i.e., either images or point clouds, which neglect the implicit semantic and geometric correlation between 2D and 3D. In this paper, we explore how the 2D modality can benefit 3D masked autoencoding, and propose Joint-MAE, a 2D-3D joint MAE framework for self-supervised 3D point cloud pre-training. Joint-MAE randomly masks an input 3D point cloud and its projected 2D images, and then reconstructs the masked information of the two modalities. For better cross-modal interaction, we construct our JointMAE by two hierarchical 2D-3D embedding modules, a joint encoder, and a joint decoder with modal-shared and model-specific decoders. On top of this, we further introduce two cross-modal strategies to boost the 3D representation learning, which are local-aligned attention mechanisms for 2D-3D semantic cues, and a cross-reconstruction loss for 2D-3D geometric constraints. By our pre-training paradigm, Joint-MAE achieves superior performance on multiple downstream tasks, e.g., 92.4% accuracy for linear SVM on ModelNet40 and 86.07% accuracy on the hardest split of ScanObjectNN.

Ziyu Guo, Renrui Zhang, Xiangyang Zhu et al.

We introduce SAM2Point, a preliminary exploration adapting Segment Anything Model 2 (SAM 2) for zero-shot and promptable 3D segmentation. SAM2Point interprets any 3D data as a series of multi-directional videos, and leverages SAM 2 for 3D-space segmentation, without further training or 2D-3D projection. Our framework supports various prompt types, including 3D points, boxes, and masks, and can generalize across diverse scenarios, such as 3D objects, indoor scenes, outdoor environments, and raw sparse LiDAR. Demonstrations on multiple 3D datasets, e.g., Objaverse, S3DIS, ScanNet, Semantic3D, and KITTI, highlight the robust generalization capabilities of SAM2Point. To our best knowledge, we present the most faithful implementation of SAM in 3D, which may serve as a starting point for future research in promptable 3D segmentation. Online Demo: https://huggingface.co/spaces/ZiyuG/SAM2Point . Code: https://github.com/ZiyuGuo99/SAM2Point .

Xiangyang Zhu, Renrui Zhang, Bowei He et al.

Large-scale pre-trained models have shown promising open-world performance for both vision and language tasks. However, their transferred capacity on 3D point clouds is still limited and only constrained to the classification task. In this paper, we first collaborate CLIP and GPT to be a unified 3D open-world learner, named as PointCLIP V2, which fully unleashes their potential for zero-shot 3D classification, segmentation, and detection. To better align 3D data with the pre-trained language knowledge, PointCLIP V2 contains two key designs. For the visual end, we prompt CLIP via a shape projection module to generate more realistic depth maps, narrowing the domain gap between projected point clouds with natural images. For the textual end, we prompt the GPT model to generate 3D-specific text as the input of CLIP's textual encoder. Without any training in 3D domains, our approach significantly surpasses PointCLIP by +42.90%, +40.44%, and +28.75% accuracy on three datasets for zero-shot 3D classification. On top of that, V2 can be extended to few-shot 3D classification, zero-shot 3D part segmentation, and 3D object detection in a simple manner, demonstrating our generalization ability for unified 3D open-world learning.

Jiaben Chen, Yichen Zhu, Dongze Lian et al. · tsinghua

Dynamic vision sensors or event cameras provide rich complementary information for video frame interpolation. Existing state-of-the-art methods follow the paradigm of combining both synthesis-based and warping networks. However, few of those methods fully respect the intrinsic characteristics of events streams. Given that event cameras only encode intensity changes and polarity rather than color intensities, estimating optical flow from events is arguably more difficult than from RGB information. We therefore propose to incorporate RGB information in an event-guided optical flow refinement strategy. Moreover, in light of the quasi-continuous nature of the time signals provided by event cameras, we propose a divide-and-conquer strategy in which event-based intermediate frame synthesis happens incrementally in multiple simplified stages rather than in a single, long stage. Extensive experiments on both synthetic and real-world datasets show that these modifications lead to more reliable and realistic intermediate frame results than previous video frame interpolation methods. Our findings underline that a careful consideration of event characteristics such as high temporal density and elevated noise benefits interpolation accuracy.

Junting Pan, Ziyi Lin, Yuying Ge et al.

Video Question Answering (VideoQA) has been significantly advanced from the scaling of recent Large Language Models (LLMs). The key idea is to convert the visual information into the language feature space so that the capacity of LLMs can be fully exploited. Existing VideoQA methods typically take two paradigms: (1) learning cross-modal alignment, and (2) using an off-the-shelf captioning model to describe the visual data. However, the first design needs costly training on many extra multi-modal data, whilst the second is further limited by limited domain generalization. To address these limitations, a simple yet effective Retrieving-to-Answer (R2A) framework is proposed.Given an input video, R2A first retrieves a set of semantically similar texts from a generic text corpus using a pre-trained multi-modal model (e.g., CLIP). With both the question and the retrieved texts, a LLM (e.g., DeBERTa) can be directly used to yield a desired answer. Without the need for cross-modal fine-tuning, R2A allows for all the key components (e.g., LLM, retrieval model, and text corpus) to plug-and-play. Extensive experiments on several VideoQA benchmarks show that despite with 1.3B parameters and no fine-tuning, our R2A can outperform the 61 times larger Flamingo-80B model even additionally trained on nearly 2.1B multi-modal data.

Bo Li, Yuanhan Zhang, Dong Guo et al.

We present LLaVA-OneVision, a family of open large multimodal models (LMMs) developed by consolidating our insights into data, models, and visual representations in the LLaVA-NeXT blog series. Our experimental results demonstrate that LLaVA-OneVision is the first single model that can simultaneously push the performance boundaries of open LMMs in three important computer vision scenarios: single-image, multi-image, and video scenarios. Importantly, the design of LLaVA-OneVision allows strong transfer learning across different modalities/scenarios, yielding new emerging capabilities. In particular, strong video understanding and cross-scenario capabilities are demonstrated through task transfer from images to videos.

Xiangyang Zhu, Renrui Zhang, Bowei He et al.

The popularity of Contrastive Language-Image Pre-training (CLIP) has propelled its application to diverse downstream vision tasks. To improve its capacity on downstream tasks, few-shot learning has become a widely-adopted technique. However, existing methods either exhibit limited performance or suffer from excessive learnable parameters. In this paper, we propose APE, an Adaptive Prior rEfinement method for CLIP's pre-trained knowledge, which achieves superior accuracy with high computational efficiency. Via a prior refinement module, we analyze the inter-class disparity in the downstream data and decouple the domain-specific knowledge from the CLIP-extracted cache model. On top of that, we introduce two model variants, a training-free APE and a training-required APE-T. We explore the trilateral affinities between the test image, prior cache model, and textual representations, and only enable a lightweight category-residual module to be trained. For the average accuracy over 11 benchmarks, both APE and APE-T attain state-of-the-art and respectively outperform the second-best by +1.59% and +1.99% under 16 shots with x30 less learnable parameters.

Ziyu Guo, Renrui Zhang, Longtian Qiu et al.

Contrastive Language-Image Pre-training (CLIP) has been shown to learn visual representations with great transferability, which achieves promising accuracy for zero-shot classification. To further improve its downstream performance, existing works propose additional learnable modules upon CLIP and fine-tune them by few-shot training sets. However, the resulting extra training cost and data requirement severely hinder the efficiency for model deployment and knowledge transfer. In this paper, we introduce a free-lunch enhancement method, CALIP, to boost CLIP's zero-shot performance via a parameter-free Attention module. Specifically, we guide visual and textual representations to interact with each other and explore cross-modal informative features via attention. As the pre-training has largely reduced the embedding distances between two modalities, we discard all learnable parameters in the attention and bidirectionally update the multi-modal features, enabling the whole process to be parameter-free and training-free. In this way, the images are blended with textual-aware signals and the text representations become visual-guided for better adaptive zero-shot alignment. We evaluate CALIP on various benchmarks of 14 datasets for both 2D image and 3D point cloud few-shot classification, showing consistent zero-shot performance improvement over CLIP. Based on that, we further insert a small number of linear layers in CALIP's attention module and verify our robustness under the few-shot settings, which also achieves leading performance compared to existing methods. Those extensive experiments demonstrate the superiority of our approach for efficient enhancement of CLIP.

Mingjie Pan, Jiaming Liu, Renrui Zhang et al.

3D occupancy prediction holds significant promise in the fields of robot perception and autonomous driving, which quantifies 3D scenes into grid cells with semantic labels. Recent works mainly utilize complete occupancy labels in 3D voxel space for supervision. However, the expensive annotation process and sometimes ambiguous labels have severely constrained the usability and scalability of 3D occupancy models. To address this, we present RenderOcc, a novel paradigm for training 3D occupancy models only using 2D labels. Specifically, we extract a NeRF-style 3D volume representation from multi-view images, and employ volume rendering techniques to establish 2D renderings, thus enabling direct 3D supervision from 2D semantics and depth labels. Additionally, we introduce an Auxiliary Ray method to tackle the issue of sparse viewpoints in autonomous driving scenarios, which leverages sequential frames to construct comprehensive 2D rendering for each object. To our best knowledge, RenderOcc is the first attempt to train multi-view 3D occupancy models only using 2D labels, reducing the dependence on costly 3D occupancy annotations. Extensive experiments demonstrate that RenderOcc achieves comparable performance to models fully supervised with 3D labels, underscoring the significance of this approach in real-world applications.

Jiaming Liu, Senqiao Yang, Peidong Jia et al.

Since real-world machine systems are running in non-stationary environments, Continual Test-Time Adaptation (CTTA) task is proposed to adapt the pre-trained model to continually changing target domains. Recently, existing methods mainly focus on model-based adaptation, which aims to leverage a self-training manner to extract the target domain knowledge. However, pseudo labels can be noisy and the updated model parameters are unreliable under dynamic data distributions, leading to error accumulation and catastrophic forgetting in the continual adaptation process. To tackle these challenges and maintain the model plasticity, we design a Visual Domain Adapter (ViDA) for CTTA, explicitly handling both domain-specific and domain-shared knowledge. Specifically, we first comprehensively explore the different domain representations of the adapters with trainable high-rank or low-rank embedding spaces. Then we inject ViDAs into the pre-trained model, which leverages high-rank and low-rank features to adapt the current domain distribution and maintain the continual domain-shared knowledge, respectively. To exploit the low-rank and high-rank ViDAs more effectively, we further propose a Homeostatic Knowledge Allotment (HKA) strategy, which adaptively combines different knowledge from each ViDA. Extensive experiments conducted on four widely used benchmarks demonstrate that our proposed method achieves state-of-the-art performance in both classification and segmentation CTTA tasks. Note that, our method can be regarded as a novel transfer paradigm for large-scale models, delivering promising results in adaptation to continually changing distributions. Project page: https://sites.google.com/view/iclr2024-vida/home.

Ruichuan An, Sihan Yang, Ziyu Guo et al.

Unified Multimodal Models (UMMs) have shown remarkable progress in visual generation. Yet, existing benchmarks predominantly assess $\textit{Crystallized Intelligence}$, which relies on recalling accumulated knowledge and learned schemas. This focus overlooks $\textit{Generative Fluid Intelligence (GFI)}$: the capacity to induce patterns, reason through constraints, and adapt to novel scenarios on the fly. To rigorously assess this capability, we introduce $\textbf{GENIUS}$ ($\textbf{GEN}$ Fluid $\textbf{I}$ntelligence Eval$\textbf{U}$ation $\textbf{S}$uite). We formalize $\textit{GFI}$ as a synthesis of three primitives. These include $\textit{Inducing Implicit Patterns}$ (e.g., inferring personalized visual preferences), $\textit{Executing Ad-hoc Constraints}$ (e.g., visualizing abstract metaphors), and $\textit{Adapting to Contextual Knowledge}$ (e.g., simulating counter-intuitive physics). Collectively, these primitives challenge models to solve problems grounded entirely in the immediate context. Our systematic evaluation of 12 representative models reveals significant performance deficits in these tasks. Crucially, our diagnostic analysis disentangles these failure modes. It demonstrates that deficits stem from limited context comprehension rather than insufficient intrinsic generative capability. To bridge this gap, we propose a training-free attention intervention strategy. Ultimately, $\textbf{GENIUS}$ establishes a rigorous standard for $\textit{GFI}$, guiding the field beyond knowledge utilization toward dynamic, general-purpose reasoning. Our dataset and code will be released at: $\href{https://github.com/arctanxarc/GENIUS}{https://github.com/arctanxarc/GENIUS}$.

Renrui Zhang, Bohao Li, Wei Zhang et al.

Few-shot classification requires deep neural networks to learn generalized representations only from limited training images, which is challenging but significant in low-data regimes. Recently, CLIP-based methods have shown promising few-shot performance benefited from the contrastive language-image pre-training. Based on this point, we question if the large-scale pre-training can alleviate the few-shot data deficiency and also assist the representation learning by the pre-learned knowledge. In this paper, we propose CoMo, a Collaboration of pre-trained Models that incorporates diverse prior knowledge from various pre-training paradigms for better few-shot learning. Our CoMo includes: CLIP's language-contrastive knowledge, DINO's vision-contrastive knowledge, and DALL-E's language-generative knowledge. Specifically, CoMo works in two aspects: few-shot data expansion and diverse knowledge ensemble. For one, we generate synthetic images via zero-shot DALL-E to enrich the few-shot training data without any manpower. For the other, we introduce a learnable Multi-Knowledge Adapter (MK-Adapter) to adaptively blend the predictions from CLIP and DINO. By such collaboration, CoMo can fully unleash the potential of different pre-training methods and unify them to perform state-of-the-art for few-shot classification. We conduct extensive experiments on 11 datasets to demonstrate the superiority and generalization ability of our approach.

Renrui Zhang, Liuhui Wang, Ziyu Guo et al.

Performances on standard 3D point cloud benchmarks have plateaued, resulting in oversized models and complex network design to make a fractional improvement. We present an alternative to enhance existing deep neural networks without any redesigning or extra parameters, termed as Spatial-Neighbor Adapter (SN-Adapter). Building on any trained 3D network, we utilize its learned encoding capability to extract features of the training dataset and summarize them as prototypical spatial knowledge. For a test point cloud, the SN-Adapter retrieves k nearest neighbors (k-NN) from the pre-constructed spatial prototypes and linearly interpolates the k-NN prediction with that of the original 3D network. By providing complementary characteristics, the proposed SN-Adapter serves as a plug-and-play module to economically improve performance in a non-parametric manner. More importantly, our SN-Adapter can be effectively generalized to various 3D tasks, including shape classification, part segmentation, and 3D object detection, demonstrating its superiority and robustness. We hope our approach could show a new perspective for point cloud analysis and facilitate future research.

Song Wen, Guian Fang, Renrui Zhang et al.

Recent advancements in text-to-image models, particularly diffusion models, have shown significant promise. However, compositional text-to-image models frequently encounter difficulties in generating high-quality images that accurately align with input texts describing multiple objects, variable attributes, and intricate spatial relationships. To address this limitation, we employ large vision-language models (LVLMs) for multi-dimensional assessment of the alignment between generated images and their corresponding input texts. Utilizing this assessment, we fine-tune the diffusion model to enhance its alignment capabilities. During the inference phase, an initial image is produced using the fine-tuned diffusion model. The LVLM is then employed to pinpoint areas of misalignment in the initial image, which are subsequently corrected using the image editing algorithm until no further misalignments are detected by the LVLM. The resultant image is consequently more closely aligned with the input text. Our experimental results validate that the proposed methodology significantly improves text-image alignment in compositional image generation, particularly with respect to object number, attribute binding, spatial relationships, and aesthetic quality.

Gengsheng Li, Jinghan He, Shijie Wang et al.

Self-play bootstraps LLM reasoning through an iterative Challenger-Solver loop: the Challenger is trained to generate questions that target the Solver's capabilities, and the Solver is optimized on the generated data to expand its reasoning skills. However, existing frameworks like R-Zero often exhibit non-sustained improvement, where early gains degrade as self-play continues. We identify a key failure mode, Diversity Illusion, where the Solver's training signals appear diverse yet collapse into recurring underlying patterns. It manifests as (1) Local Diversity Illusion, where diversity is enforced only within-batch, inducing cross-iteration mode cycling; and (2) Surface Diversity Illusion, where questions vary superficially but require near-identical reasoning skills. To mitigate them, we propose R-Diverse with two aligned innovations: Memory-Augmented Penalty (MAP), which uses a persistent memory bank to discourage recycling across iterations, and Skill-Aware Measurement (SAM), which evaluates diversity by the reasoning skills exercised rather than surface variation of questions. Across 10 math and general reasoning benchmarks, R-Diverse sustains gains over more iterations and consistently outperforms prior self-play methods. Code is available at https://github.com/Gengsheng-Li/R-Diverse.

Bowei He, Xu He, Renrui Zhang et al.

With the continuous increase of users and items, conventional recommender systems trained on static datasets can hardly adapt to changing environments. The high-throughput data requires the model to be updated in a timely manner for capturing the user interest dynamics, which leads to the emergence of streaming recommender systems. Due to the prevalence of deep learning-based recommender systems, the embedding layer is widely adopted to represent the characteristics of users, items, and other features in low-dimensional vectors. However, it has been proved that setting an identical and static embedding size is sub-optimal in terms of recommendation performance and memory cost, especially for streaming recommendations. To tackle this problem, we first rethink the streaming model update process and model the dynamic embedding size search as a bandit problem. Then, we analyze and quantify the factors that influence the optimal embedding sizes from the statistics perspective. Based on this, we propose the \textbf{D}ynamic \textbf{E}mbedding \textbf{S}ize \textbf{S}earch (\textbf{DESS}) method to minimize the embedding size selection regret on both user and item sides in a non-stationary manner. Theoretically, we obtain a sublinear regret upper bound superior to previous methods. Empirical results across two recommendation tasks on four public datasets also demonstrate that our approach can achieve better streaming recommendation performance with lower memory cost and higher time efficiency.

Yuanhong Zheng, Ruichuan An, Xiaopeng Lin et al.

Human cognition of new concepts is inherently a streaming process: we continuously recognize new objects or identities and update our memories over time. However, current multimodal personalization methods are largely limited to static images or offline videos. This disconnects continuous visual input from instant real-world feedback, limiting their ability to provide the real-time, interactive personalized responses essential for future AI assistants. To bridge this gap, we first propose and formally define the novel task of Personalized Streaming Video Understanding (PSVU). To facilitate research in this new direction, we introduce PEARL-Bench, the first comprehensive benchmark designed specifically to evaluate this challenging setting. It evaluates a model's ability to respond to personalized concepts at exact timestamps under two modes: (1) Frame-level, focusing on a specific person or object in discrete frames, and (2) a novel Video-level, focusing on personalized actions unfolding across continuous frames. PEARL-Bench comprises 132 unique videos and 2,173 fine-grained annotations with precise timestamps. Concept diversity and annotation quality are strictly ensured through a combined pipeline of automated generation and human verification. To tackle this challenging new setting, we further propose PEARL, a plug-and-play, training-free strategy that serves as a strong baseline. Extensive evaluations across 8 offline and online models demonstrate that PEARL achieves state-of-the-art performance. Notably, it brings consistent PSVU improvements when applied to 3 distinct architectures, proving to be a highly effective and robust strategy. We hope this work advances vision-language model (VLM) personalization and inspires further research into streaming personalized AI assistants. Code is available at https://github.com/Yuanhong-Zheng/PEARL.

Zhuoyang Liu, Jiaming Liu, Jiadong Xu et al.

Vision-language-action models (VLAs) have shown generalization capabilities in robotic manipulation tasks by inheriting from vision-language models (VLMs) and learning action generation. Most VLA models focus on interpreting vision and language to generate actions, whereas robots must perceive and interact within the spatial-physical world. This gap highlights the need for a comprehensive understanding of robotic-specific multisensory information, which is crucial for achieving complex and contact-rich control. To this end, we introduce a multisensory language-action (MLA) model that collaboratively perceives heterogeneous sensory modalities and predicts future multisensory objectives to facilitate physical world modeling. Specifically, to enhance perceptual representations, we propose an encoder-free multimodal alignment scheme that innovatively repurposes the large language model itself as a perception module, directly interpreting multimodal cues by aligning 2D images, 3D point clouds, and tactile tokens through positional correspondence. To further enhance MLA's understanding of physical dynamics, we design a future multisensory generation post-training strategy that enables MLA to reason about semantic, geometric, and interaction information, providing more robust conditions for action generation. For evaluation, the MLA model outperforms the previous state-of-the-art 2D and 3D VLA methods by 12% and 24% in complex, contact-rich real-world tasks, respectively, while also demonstrating improved generalization to unseen configurations.

Zhuoyang Liu, Jiaming Liu, Hao Chen et al.

Vision-Language-Action (VLA) models have recently shown strong generalization, with some approaches seeking to explicitly generate linguistic reasoning traces or predict future observations prior to execution. However, explicit reasoning typically incurs non-negligible inference latency, which constrains the temporal resolution required for robotic manipulation. Moreover, such reasoning is confined to the linguistic space, imposing a representational bottleneck that struggles to faithfully capture ineffable physical attributes. To mitigate these limitations, we propose LaST$_0$, a framework that enables efficient reasoning before acting through a Latent Spatio-Temporal Chain-of-Thought (CoT), capturing fine-grained physical and robotic dynamics that are often difficult to verbalize. Specifically, we introduce a token-efficient latent CoT space that models future visual dynamics, 3D structural information, and robot proprioceptive states, and further extends these representations across time to enable temporally consistent implicit reasoning trajectories. Furthermore, LaST$_0$ adopts a dual-system architecture implemented via a Mixture-of-Transformers design, where a reasoning expert conducts low-frequency latent inference and an acting expert generates high-frequency actions conditioned on robotics-oriented latent representations. To facilitate coordination, LaST$_0$ is trained with heterogeneous operation frequencies, enabling adaptive switching during deployment. Across 10 real-world tasks spanning tabletop, mobile, and dexterous hand manipulation, LaST$_0$ improves mean success rates by 13%, 14% and 14% over prior SOTA VLA methods, respectively.

Tianyi Xiong, Yi Ge, Ming Li et al.

Large multimodal models (LMMs) are increasingly adopted as judges in multimodal evaluation systems due to their strong instruction following and consistency with human preferences. However, their ability to follow diverse, fine-grained evaluation criteria remains underexplored. We develop Multi-Crit, a benchmark for evaluating multimodal judges on their capacity to follow pluralistic criteria and produce reliable criterion-level judgments. Covering both open-ended generation and verifiable reasoning tasks, Multi-Crit is built through a rigorous data curation pipeline that gathers challenging response pairs with multi-criterion human annotations. It further introduces three novel metrics for systematically assessing pluralistic adherence, criterion-switching flexibility, and the ability to recognize criterion-level preference conflicts. Comprehensive analysis of 25 LMMs reveals that 1) proprietary models still struggle to maintain consistent adherence to pluralistic criteria--especially in open-ended evaluation; 2) open-source models lag further behind in flexibly following diverse criteria; and 3) critic fine-tuning with holistic judgment signals enhances visual grounding but fails to generalize to pluralistic criterion-level judgment. Additional analyses on reasoning fine-tuning, test-time scaling, and boundary consistency between open-source and proprietary models further probe the limits of current multimodal judges. As a pioneering study, Multi-Crit lays the foundation for building reliable and steerable multimodal AI evaluation.

Shilin Yan, Xiaohao Xu, Renrui Zhang et al.

Panoramic videos contain richer spatial information and have attracted tremendous amounts of attention due to their exceptional experience in some fields such as autonomous driving and virtual reality. However, existing datasets for video segmentation only focus on conventional planar images. To address the challenge, in this paper, we present a panoramic video dataset, PanoVOS. The dataset provides 150 videos with high video resolutions and diverse motions. To quantify the domain gap between 2D planar videos and panoramic videos, we evaluate 15 off-the-shelf video object segmentation (VOS) models on PanoVOS. Through error analysis, we found that all of them fail to tackle pixel-level content discontinues of panoramic videos. Thus, we present a Panoramic Space Consistency Transformer (PSCFormer), which can effectively utilize the semantic boundary information of the previous frame for pixel-level matching with the current frame. Extensive experiments demonstrate that compared with the previous SOTA models, our PSCFormer network exhibits a great advantage in terms of segmentation results under the panoramic setting. Our dataset poses new challenges in panoramic VOS and we hope that our PanoVOS can advance the development of panoramic segmentation/tracking.

Kaining Ying, Fanqing Meng, Jin Wang et al.

Large Vision-Language Models (LVLMs) show significant strides in general-purpose multimodal applications such as visual dialogue and embodied navigation. However, existing multimodal evaluation benchmarks cover a limited number of multimodal tasks testing rudimentary capabilities, falling short in tracking LVLM development. In this study, we present MMT-Bench, a comprehensive benchmark designed to assess LVLMs across massive multimodal tasks requiring expert knowledge and deliberate visual recognition, localization, reasoning, and planning. MMT-Bench comprises $31,325$ meticulously curated multi-choice visual questions from various multimodal scenarios such as vehicle driving and embodied navigation, covering $32$ core meta-tasks and $162$ subtasks in multimodal understanding. Due to its extensive task coverage, MMT-Bench enables the evaluation of LVLMs using a task map, facilitating the discovery of in- and out-of-domain tasks. Evaluation results involving $30$ LVLMs such as the proprietary GPT-4V, GeminiProVision, and open-sourced InternVL-Chat, underscore the significant challenges posed by MMT-Bench. We anticipate that MMT-Bench will inspire the community to develop next-generation multimodal foundation models aimed at achieving general-purpose multimodal intelligence.

Zijun Shen, Sihan Yang, Ruichuan An et al.

Unified Multimodal Models (UMMs) excel in general tasks but struggle to bridge the gap between personalized understanding and generation. Prior works largely rely on implicit token-level alignment via supervised fine-tuning, which fails to fully capture the potential synergy between comprehension and creation. In this work, we propose Sync-R1, an end-to-end reinforcement learning framework that jointly optimizes personalized understanding and generation within a single, explicit reasoning loop. Through this unified feedback process, Sync-R1 enables personalized comprehension to guide content creation, while the resulting generation quality reciprocally refines understanding within an integrated reward landscape. To efficiently orchestrate this dual-task synergy, we introduce Sync-GRPO, a reinforcement learning method utilizing an ensemble reward system. Furthermore, we propose Dynamic Group Scaling (DGS), which adaptively filters low-potential trajectories to reduce gradient variance and accelerate convergence. To better reflect real-world complexity, we introduce UnifyBench++, featuring denser textual descriptions and richer user contexts. Experimental results demonstrate that Sync-R1 achieves state-of-the-art performance, showcasing superior cross-task reasoning and robust personalization without requiring complex cold-start procedures. The code and the UnifyBench++ dataset will be released at: https://github.com/arctanxarc/UniCTokens.