AlignBench: Benchmarking Chinese Alignment of Large Language Models

Alignment has become a critical step for instruction-tuned Large Language Models (LLMs) to become helpful assistants. However, effective evaluation of alignment for emerging Chinese LLMs is still significantly lacking, calling for real-scenario grounded, open-ended, challenging and automatic evaluations tailored for alignment. To fill in this gap, we introduce AlignBench, a comprehensive multi-dimensional benchmark for evaluating LLMs' alignment in Chinese. Equipped with a human-in-the-loop data curation pipeline, our benchmark employs a rule-calibrated multi-dimensional LLM-as-Judge with Chain-of-Thought to generate explanations and final ratings as evaluations, ensuring high reliability and interpretability. Furthermore, we developed a dedicated companion evaluator LLM -- CritiqueLLM, which recovers 95\% of GPT-4's evaluation ability and will be provided via public APIs to researchers for evaluation of alignment in Chinese LLMs. All evaluation codes, data, and LLM generations are available at \url{https://github.com/THUDM/AlignBench}.

HandRefiner: Refining Malformed Hands in Generated Images by Diffusion-based Conditional Inpainting

Diffusion models have achieved remarkable success in generating realistic images but suffer from generating accurate human hands, such as incorrect finger counts or irregular shapes. This difficulty arises from the complex task of learning the physical structure and pose of hands from training images, which involves extensive deformations and occlusions. For correct hand generation, our paper introduces a lightweight post-processing solution called $\textbf{HandRefiner}$. HandRefiner employs a conditional inpainting approach to rectify malformed hands while leaving other parts of the image untouched. We leverage the hand mesh reconstruction model that consistently adheres to the correct number of fingers and hand shape, while also being capable of fitting the desired hand pose in the generated image. Given a generated failed image due to malformed hands, we utilize ControlNet modules to re-inject such correct hand information. Additionally, we uncover a phase transition phenomenon within ControlNet as we vary the control strength. It enables us to take advantage of more readily available synthetic data without suffering from the domain gap between realistic and synthetic hands. Experiments demonstrate that HandRefiner can significantly improve the generation quality quantitatively and qualitatively. The code is available at https://github.com/wenquanlu/HandRefiner .

ArGue: Attribute-Guided Prompt Tuning for Vision-Language Models

Although soft prompt tuning is effective in efficiently adapting Vision-Language (V&L) models for downstream tasks, it shows limitations in dealing with distribution shifts. We address this issue with Attribute-Guided Prompt Tuning (ArGue), making three key contributions. 1) In contrast to the conventional approach of directly appending soft prompts preceding class names, we align the model with primitive visual attributes generated by Large Language Models (LLMs). We posit that a model's ability to express high confidence in these attributes signifies its capacity to discern the correct class rationales. 2) We introduce attribute sampling to eliminate disadvantageous attributes, thus only semantically meaningful attributes are preserved. 3) We propose negative prompting, explicitly enumerating class-agnostic attributes to activate spurious correlations and encourage the model to generate highly orthogonal probability distributions in relation to these negative features. In experiments, our method significantly outperforms current state-of-the-art prompt tuning methods on both novel class prediction and out-of-distribution generalization tasks.

Low-Complexity Joint Beamforming for RIS-Assisted MU-MISO Systems Based on Model-Driven Deep Learning

Reconfigurable intelligent surfaces (RIS) can improve signal propagation environments by adjusting the phase of the incident signal. However, optimizing the phase shifts jointly with the beamforming vector at the access point is challenging due to the non-convex objective function and constraints. In this study, we propose an algorithm based on weighted minimum mean square error optimization and power iteration to maximize the weighted sum rate (WSR) of a RIS-assisted downlink multi-user multiple-input single-output system. To further improve performance, a model-driven deep learning (DL) approach is designed, where trainable variables and graph neural networks are introduced to accelerate the convergence of the proposed algorithm. We also extend the proposed method to include beamforming with imperfect channel state information and derive a two-timescale stochastic optimization algorithm. Simulation results show that the proposed algorithm outperforms state-of-the-art algorithms in terms of complexity and WSR. Specifically, the model-driven DL approach has a runtime that is approximately 3% of the state-of-the-art algorithm to achieve the same performance. Additionally, the proposed algorithm with 2-bit phase shifters outperforms the compared algorithm with continuous phase shift.

DA-STC: Domain Adaptive Video Semantic Segmentation via Spatio-Temporal Consistency

Video semantic segmentation is a pivotal aspect of video representation learning. However, significant domain shifts present a challenge in effectively learning invariant spatio-temporal features across the labeled source domain and unlabeled target domain for video semantic segmentation. To solve the challenge, we propose a novel DA-STC method for domain adaptive video semantic segmentation, which incorporates a bidirectional multi-level spatio-temporal fusion module and a category-aware spatio-temporal feature alignment module to facilitate consistent learning for domain-invariant features. Firstly, we perform bidirectional spatio-temporal fusion at the image sequence level and shallow feature level, leading to the construction of two fused intermediate video domains. This prompts the video semantic segmentation model to consistently learn spatio-temporal features of shared patch sequences which are influenced by domain-specific contexts, thereby mitigating the feature gap between the source and target domain. Secondly, we propose a category-aware feature alignment module to promote the consistency of spatio-temporal features, facilitating adaptation to the target domain. Specifically, we adaptively aggregate the domain-specific deep features of each category along spatio-temporal dimensions, which are further constrained to achieve cross-domain intra-class feature alignment and inter-class feature separation. Extensive experiments demonstrate the effectiveness of our method, which achieves state-of-the-art mIOUs on multiple challenging benchmarks. Furthermore, we extend the proposed DA-STC to the image domain, where it also exhibits superior performance for domain adaptive semantic segmentation. The source code and models will be made available at \url{https://github.com/ZHE-SAPI/DA-STC}.

An LLM-free Multi-dimensional Benchmark for MLLMs Hallucination Evaluation

Despite making significant progress in multi-modal tasks, current Multi-modal Large Language Models (MLLMs) encounter the significant challenge of hallucination, which may lead to harmful consequences. Therefore, evaluating MLLMs' hallucinations is becoming increasingly important in model improvement and practical application deployment. Previous works are limited in high evaluation costs (e.g., relying on humans or advanced LLMs) and insufficient evaluation dimensions (e.g., types of hallucination and task). In this paper, we propose an LLM-free multi-dimensional benchmark AMBER, which can be used to evaluate both generative task and discriminative task including object existence, object attribute and object relation hallucination. Based on AMBER, we design a low-cost and efficient evaluation pipeline. Additionally, we conduct a comprehensive evaluation and detailed analysis of mainstream MLLMs including GPT-4V(ision), and also give guideline suggestions for mitigating hallucinations. The data and code of AMBER are available at https://github.com/junyangwang0410/AMBER.

IMPUS: Image Morphing with Perceptually-Uniform Sampling Using Diffusion Models

We present a diffusion-based image morphing approach with perceptually-uniform sampling (IMPUS) that produces smooth, direct, and realistic interpolations given an image pair. A latent diffusion model has distinct conditional distributions and data embeddings for each of the two images, especially when they are from different classes. To bridge this gap, we interpolate in the locally linear and continuous text embedding space and Gaussian latent space. We first optimize the endpoint text embeddings and then map the images to the latent space using a probability flow ODE. Unlike existing work that takes an indirect morphing path, we show that the model adaptation yields a direct path and suppresses ghosting artifacts in the interpolated images. To achieve this, we propose an adaptive bottleneck constraint based on a novel relative perceptual path diversity score that automatically controls the bottleneck size and balances the diversity along the path with its directness. We also propose a perceptually-uniform sampling technique that enables visually smooth changes between the interpolated images. Extensive experiments validate that our IMPUS can achieve smooth, direct, and realistic image morphing and be applied to other image generation tasks.

PT-Tuning: Bridging the Gap between Time Series Masked Reconstruction and Forecasting via Prompt Token Tuning

Self-supervised learning has been actively studied in time series domain recently, especially for masked reconstruction. Most of these methods follow the "Pre-training + Fine-tuning" paradigm in which a new decoder replaces the pre-trained decoder to fit for a specific downstream task, leading to inconsistency of upstream and downstream tasks. In this paper, we first point out that the unification of task objectives and adaptation for task difficulty are critical for bridging the gap between time series masked reconstruction and forecasting. By reserving the pre-trained mask token during fine-tuning stage, the forecasting task can be taken as a special case of masked reconstruction, where the future values are masked and reconstructed based on history values. It guarantees the consistency of task objectives but there is still a gap in task difficulty. Because masked reconstruction can utilize contextual information while forecasting can only use historical information to reconstruct. To further mitigate the existed gap, we propose a simple yet effective prompt token tuning (PT-Tuning) paradigm, in which all pre-trained parameters are frozen and only a few trainable prompt tokens are added to extended mask tokens in element-wise manner. Extensive experiments on real-world datasets demonstrate the superiority of our proposed paradigm with state-of-the-art performance compared to representation learning and end-to-end supervised forecasting methods.

Multimodal Variational Auto-encoder based Audio-Visual Segmentation

We propose an Explicit Conditional Multimodal Variational Auto-Encoder (ECMVAE) for audio-visual segmentation (AVS), aiming to segment sound sources in the video sequence. Existing AVS methods focus on implicit feature fusion strategies, where models are trained to fit the discrete samples in the dataset. With a limited and less diverse dataset, the resulting performance is usually unsatisfactory. In contrast, we address this problem from an effective representation learning perspective, aiming to model the contribution of each modality explicitly. Specifically, we find that audio contains critical category information of the sound producers, and visual data provides candidate sound producer(s). Their shared information corresponds to the target sound producer(s) shown in the visual data. In this case, cross-modal shared representation learning is especially important for AVS. To achieve this, our ECMVAE factorizes the representations of each modality with a modality-shared representation and a modality-specific representation. An orthogonality constraint is applied between the shared and specific representations to maintain the exclusive attribute of the factorized latent code. Further, a mutual information maximization regularizer is introduced to achieve extensive exploration of each modality. Quantitative and qualitative evaluations on the AVSBench demonstrate the effectiveness of our approach, leading to a new state-of-the-art for AVS, with a 3.84 mIOU performance leap on the challenging MS3 subset for multiple sound source segmentation.

RPEFlow: Multimodal Fusion of RGB-PointCloud-Event for Joint Optical Flow and Scene Flow Estimation

Recently, the RGB images and point clouds fusion methods have been proposed to jointly estimate 2D optical flow and 3D scene flow. However, as both conventional RGB cameras and LiDAR sensors adopt a frame-based data acquisition mechanism, their performance is limited by the fixed low sampling rates, especially in highly-dynamic scenes. By contrast, the event camera can asynchronously capture the intensity changes with a very high temporal resolution, providing complementary dynamic information of the observed scenes. In this paper, we incorporate RGB images, Point clouds and Events for joint optical flow and scene flow estimation with our proposed multi-stage multimodal fusion model, RPEFlow. First, we present an attention fusion module with a cross-attention mechanism to implicitly explore the internal cross-modal correlation for 2D and 3D branches, respectively. Second, we introduce a mutual information regularization term to explicitly model the complementary information of three modalities for effective multimodal feature learning. We also contribute a new synthetic dataset to advocate further research. Experiments on both synthetic and real datasets show that our model outperforms the existing state-of-the-art by a wide margin. Code and dataset is available at https://npucvr.github.io/RPEFlow.