We introduce the Region Encoder Network (REN), a fast and effective model for generating region-based image representations using point prompts. Recent methods combine class-agnostic segmenters (e.g., SAM) with patch-based image encoders (e.g., DINO) to produce compact and effective region representations, but they suffer from high computational cost due to the segmentation step. REN bypasses this bottleneck using a lightweight module that directly generates region tokens, enabling 60x faster token generation with 35x less memory, while also improving token quality. It uses a few cross-attention blocks that take point prompts as queries and features from a patch-based image encoder as keys and values to produce region tokens that correspond to the prompted objects. We train REN with three popular encoders-DINO, DINOv2, and OpenCLIP-and show that it can be extended to other encoders without dedicated training. We evaluate REN on semantic segmentation and retrieval tasks, where it consistently outperforms the original encoders in both performance and compactness, and matches or exceeds SAM-based region methods while being significantly faster. Notably, REN achieves state-of-the-art results on the challenging Ego4D VQ2D benchmark and outperforms proprietary LMMs on Visual Haystacks' single-needle challenge. Code and models are available at: https://github.com/savya08/REN.
기존 RL 사후학습 방법들이 현재 배치의 보상으로만 정책을 업데이트하는 개루프(open-loop) 구조의 한계를 지적하며, 반복 간 성능 향상 자체를 명시적 목표로 삼는 PIRL(Policy Improvement Reinforcement Learning) 프레임워크와 이를 기반으로 한 폐루프(closed-loop) 플러그인 알고리즘 PIPO를 제안한다. PIPO는 슬라이딩 윈도우 방식의 과거 성능 앵커와 비교해 업데이트 효과를 검증하고, PPO·GRPO·SDPO 등 다양한 정책 최적화 계열에서 수학적 추론, 코드, 도구 사용 태스크 전반에 걸쳐 일관된 성능 향상을 달성했다.
설명 가능한 AI(XAI)는 이제 NLP의 주요 주제가 되었으나, 아랍어 NLP는 서로 연결된 세 가지 측면에서 여전히 설명이 부족한 상태이다. 첫째, 방법 격차가 있다. 아랍어 XAI는 LIME, SHAP, 어텐션 시각화, saliency와 같이 소수의 사후(post-hoc) 기법에 크게 의존하는 반면, 더 넓은 NLP XAI 분야에서는 진단, 반사실(counterfactual), 프로빙, 근거(rationale) 기반, 인간 중심 방법 등 더 풍부한 방법을 제공한다. 둘째, 과제 격차가 있다. 기존 아랍어 XAI 연구는 분류 과제, 특히 감성 분석, 혐오/모욕 언어 탐지, 가짜 뉴스, 스팸에 집중되어 있으며, 생성, 검색, 번역, 요약, 구조적 예측, 대화에 대한 다룸은 상대적으로 미흡하다. 셋째, 언어적 격차가 있다. 많은 설명이 영향력 있는 토큰을 식별하지만, 형태론, 접어(clitic), 방언 변이, 이중언어 상용(diglossia), 표기 모호성, 발음 부호(diacritics), 코드 스위칭, 개체명, 문화적 지칭, 또는 고전 및 종교적 문체와 같은 아랍어 고유의 현상은 거의 설명하지 못한다. 이 비판적 구조적 서베이는 텍스트, 음성, 멀티모달 환경에 걸친 아랍어 XAI 관련 검토 문헌을 종합한다. 우리는 아랍어 NLP에 필요한 것이 모델 결정에 대한 설명만이 아니라, 언어적·문화적·사회기술적 대상으로서의 아랍어에 충실한 설명이라고 주장한다. 우리는 과제, 방법, 언어 단위, 변종, 목표, 평가 관행에 대한 분류 체계를 제시하고, 언어학적으로 그라운딩된 아랍어 XAI를 위한 연구 아젠다를 제안한다.
확산 모델은 고품질 이미지를 생성하지만 얼마나 잘 일반화하는지 측정하기는 어렵다. 텍스트-이미지 모델은 사람의 선호와 프롬프트 준수 벤치마크로, 클래스 조건부 모델은 프레셰 인셉션 거리(FID)와 인셉션 점수(IS) 같은 분포 수준 평가지표로 주로 평가한다. 그러나 이 방법으로는 모델이 기본 시각 능력을 배웠는지, 주로 학습 패턴을 따라 하는지 구별할 수 없다. 확산 모델이 크기, 위치, 회전, 개수, 색, 모양 같은 기본 시각 능력과 그 조합을 어떻게 학습하고 일반화하는지 연구하기 위해 통제된 실험 프레임워크 Visual Generative Lab(VGL)을 소개한다. 명시적인 규칙으로 능력별 합성 학습 데이터를 생성해 과제에 맞는 규칙 기반 평가지표로 일반화를 정밀하게 측정한다. 이 데이터에서 모델을 처음부터 학습한 결과 세 가지 양상을 확인했다. (1) 기본 설계에서는 외삽에 한계가 있었다. 크기와 위치는 학습 범위에서 몇 픽셀만 벗어나도 성능이 떨어졌다. 범위를 벗어난 개수에서는 정확도가 약 50%였고, 회전 외삽의 정확도는 시드마다 크게 달랐다. (2) 조합적 일반화에서는 데이터셋 크기보다 서로 다른 기본 시각 능력 조합의 수가 더 중요했다. (3) 위치를 제외하면 각 능력은 대체로 서로 독립적으로 실패했다. 코드와 데이터 생성 스크립트는 https://github.com/AmishSethi/visual-generative-lab 에 공개되어 있다.
Human pose estimation - the process of recognizing human keypoints in a given image - is one of the most important tasks in computer vision and has a wide range of applications including movement diagnostics, surveillance, or self-driving vehicle. The accuracy of human keypoint prediction is increasingly improved thanks to the burgeoning development of deep learning. Most existing methods solved human pose estimation by generating heatmaps in which the ith heatmap indicates the location confidence of the ith keypoint. In this paper, we introduce novel network structures referred to as multi-resolution representation learning for human keypoint prediction. At different resolutions in the learning process, our networks branch off and use extra layers to learn heatmap generation. We firstly consider the architectures for generating the multi-resolution heatmaps after obtaining the lowest-resolution feature maps. Our second approach allows learning during the process of feature extraction in which the heatmaps are generated at each resolution of the feature extractor. The first and second approaches are referred to as multi-resolution heatmap learning and multi-resolution feature map learning respectively. Our architectures are simple yet effective, achieving good performance. We conducted experiments on two common benchmarks for human pose estimation: MSCOCO and MPII dataset. The code is made publicly available at https://github.com/tqtrunghnvn/SimMRPose.
Recent research finds CNN models for image classification demonstrate overlapped adversarial vulnerabilities: adversarial attacks can mislead CNN models with small perturbations, which can effectively transfer between different models trained on the same dataset. Adversarial training, as a general robustness improvement technique, eliminates the vulnerability in a single model by forcing it to learn robust features. The process is hard, often requires models with large capacity, and suffers from significant loss on clean data accuracy. Alternatively, ensemble methods are proposed to induce sub-models with diverse outputs against a transfer adversarial example, making the ensemble robust against transfer attacks even if each sub-model is individually non-robust. Only small clean accuracy drop is observed in the process. However, previous ensemble training methods are not efficacious in inducing such diversity and thus ineffective on reaching robust ensemble. We propose DVERGE, which isolates the adversarial vulnerability in each sub-model by distilling non-robust features, and diversifies the adversarial vulnerability to induce diverse outputs against a transfer attack. The novel diversity metric and training procedure enables DVERGE to achieve higher robustness against transfer attacks comparing to previous ensemble methods, and enables the improved robustness when more sub-models are added to the ensemble. The code of this work is available at https://github.com/zjysteven/DVERGE
Training autonomous agents able to generalize to multiple tasks is a key target of Deep Reinforcement Learning (DRL) research. In parallel to improving DRL algorithms themselves, Automatic Curriculum Learning (ACL) study how teacher algorithms can train DRL agents more efficiently by adapting task selection to their evolving abilities. While multiple standard benchmarks exist to compare DRL agents, there is currently no such thing for ACL algorithms. Thus, comparing existing approaches is difficult, as too many experimental parameters differ from paper to paper. In this work, we identify several key challenges faced by ACL algorithms. Based on these, we present TeachMyAgent (TA), a benchmark of current ACL algorithms leveraging procedural task generation. It includes 1) challenge-specific unit-tests using variants of a procedural Box2D bipedal walker environment, and 2) a new procedural Parkour environment combining most ACL challenges, making it ideal for global performance assessment. We then use TeachMyAgent to conduct a comparative study of representative existing approaches, showcasing the competitiveness of some ACL algorithms that do not use expert knowledge. We also show that the Parkour environment remains an open problem. We open-source our environments, all studied ACL algorithms (collected from open-source code or re-implemented), and DRL students in a Python package available at https://github.com/flowersteam/TeachMyAgent.
In tissue characterization and cancer diagnostics, multimodal imaging has emerged as a powerful technique. Thanks to computational advances, large datasets can be exploited to discover patterns in pathologies and improve diagnosis. However, this requires efficient and scalable image retrieval methods. Cross-modality image retrieval is particularly challenging, since images of similar (or even the same) content captured by different modalities might share few common structures. We propose a new application-independent content-based image retrieval (CBIR) system for reverse (sub-)image search across modalities, which combines deep learning to generate representations (embedding the different modalities in a common space) with classical feature extraction and bag-of-words models for efficient and reliable retrieval. We illustrate its advantages through a replacement study, exploring a number of feature extractors and learned representations, as well as through comparison to recent (cross-modality) CBIR methods. For the task of (sub-)image retrieval on a (publicly available) dataset of brightfield and second harmonic generation microscopy images, the results show that our approach is superior to all tested alternatives. We discuss the shortcomings of the compared methods and observe the importance of equivariance and invariance properties of the learned representations and feature extractors in the CBIR pipeline. Code is available at: \url{https://github.com/MIDA-group/CrossModal_ImgRetrieval}.
In-vitro tests are an alternative to animal testing for the toxicity of medical devices. Detecting cells as a first step, a cell expert evaluates the growth of cells according to cytotoxicity grade under the microscope. Thus, human fatigue plays a role in error making, making the use of deep learning appealing. Due to the high cost of training data annotation, an approach without manual annotation is needed. We propose Seamless Iterative Semi-Supervised correction of Imperfect labels (SISSI), a new method for training object detection models with noisy and missing annotations in a semi-supervised fashion. Our network learns from noisy labels generated with simple image processing algorithms, which are iteratively corrected during self-training. Due to the nature of missing bounding boxes in the pseudo labels, which would negatively affect the training, we propose to train on dynamically generated synthetic-like images using seamless cloning. Our method successfully provides an adaptive early learning correction technique for object detection. The combination of early learning correction that has been applied in classification and semantic segmentation before and synthetic-like image generation proves to be more effective than the usual semi-supervised approach by > 15% AP and > 20% AR across three different readers. Our code is available at https://github.com/marwankefah/SISSI.
The next generation of Internet services, such as Metaverse, rely on mixed reality (MR) technology to provide immersive user experiences. However, the limited computation power of MR headset-mounted devices (HMDs) hinders the deployment of such services. Therefore, we propose an efficient information sharing scheme based on full-duplex device-to-device (D2D) semantic communications to address this issue. Our approach enables users to avoid heavy and repetitive computational tasks, such as artificial intelligence-generated content (AIGC) in the view images of all MR users. Specifically, a user can transmit the generated content and semantic information extracted from their view image to nearby users, who can then use this information to obtain the spatial matching of computation results under their view images. We analyze the performance of full-duplex D2D communications, including the achievable rate and bit error probability, by using generalized small-scale fading models. To facilitate semantic information sharing among users, we design a contract theoretic AI-generated incentive mechanism. The proposed diffusion model generates the optimal contract design, outperforming two deep reinforcement learning algorithms, i.e., proximal policy optimization and soft actor-critic algorithms. Our numerical analysis experiment proves the effectiveness of our proposed methods. The code for this paper is available at https://github.com/HongyangDu/SemSharing
Similar to the majority of deep learning applications, diagnosing skin diseases using computer vision and deep learning often requires a large volume of data. However, obtaining sufficient data for particular types of facial skin conditions can be difficult due to privacy concerns. As a result, conditions like Rosacea are often understudied in computer-aided diagnosis. The limited availability of data for facial skin conditions has led to the investigation of alternative methods for computer-aided diagnosis. In recent years, Generative Adversarial Networks (GANs), mainly variants of StyleGANs, have demonstrated promising results in generating synthetic facial images. In this study, for the first time, a small dataset of Rosacea with 300 full-face images is utilized to further investigate the possibility of generating synthetic data. The preliminary experiments show how fine-tuning the model and varying experimental settings significantly affect the fidelity of the Rosacea features. It is demonstrated that $R_1$ Regularization strength helps achieve high-fidelity details. Additionally, this study presents qualitative evaluations of synthetic/generated faces by expert dermatologists and non-specialist participants. The quantitative evaluation is presented using a few validation metric(s). Furthermore a number of limitations and future directions are discussed. Code and generated dataset are available at: \url{https://github.com/thinkercache/stylegan2-ada-pytorch}
The ability to understand visual concepts and replicate and compose these concepts from images is a central goal for computer vision. Recent advances in text-to-image (T2I) models have lead to high definition and realistic image quality generation by learning from large databases of images and their descriptions. However, the evaluation of T2I models has focused on photorealism and limited qualitative measures of visual understanding. To quantify the ability of T2I models in learning and synthesizing novel visual concepts (a.k.a. personalized T2I), we introduce ConceptBed, a large-scale dataset that consists of 284 unique visual concepts, and 33K composite text prompts. Along with the dataset, we propose an evaluation metric, Concept Confidence Deviation (CCD), that uses the confidence of oracle concept classifiers to measure the alignment between concepts generated by T2I generators and concepts contained in target images. We evaluate visual concepts that are either objects, attributes, or styles, and also evaluate four dimensions of compositionality: counting, attributes, relations, and actions. Our human study shows that CCD is highly correlated with human understanding of concepts. Our results point to a trade-off between learning the concepts and preserving the compositionality which existing approaches struggle to overcome. The data, code, and interactive demo is available at: https://conceptbed.github.io/
The latest trends in the research field of single-view human reconstruction devote to learning deep implicit functions constrained by explicit body shape priors. Despite the remarkable performance improvements compared with traditional processing pipelines, existing learning approaches still show different aspects of limitations in terms of flexibility, generalizability, robustness, and/or representation capability. To comprehensively address the above issues, in this paper, we investigate an explicit point-based human reconstruction framework called HaP, which adopts point clouds as the intermediate representation of the target geometric structure. Technically, our approach is featured by fully-explicit point cloud estimation, manipulation, generation, and refinement in the 3D geometric space, instead of an implicit learning process that can be ambiguous and less controllable. The overall workflow is carefully organized with dedicated designs of the corresponding specialized learning components as well as processing procedures. Extensive experiments demonstrate that our framework achieves quantitative performance improvements of 20% to 40% over current state-of-the-art methods, and better qualitative results. Our promising results may indicate a paradigm rollback to the fully-explicit and geometry-centric algorithm design, which enables to exploit various powerful point cloud modeling architectures and processing techniques. We will make our code and data publicly available at https://github.com/yztang4/HaP.
Modern Large Vision-Language Models (LVLMs) enjoy the same vision vocabulary -- CLIP, which can cover most common vision tasks. However, for some special vision task that needs dense and fine-grained vision perception, e.g., document-level OCR or chart understanding, especially in non-English scenarios, the CLIP-style vocabulary may encounter low efficiency in tokenizing the vision knowledge and even suffer out-of-vocabulary problem. Accordingly, we propose Vary, an efficient and effective method to scale up the vision vocabulary of LVLMs. The procedures of Vary are naturally divided into two folds: the generation and integration of a new vision vocabulary. In the first phase, we devise a vocabulary network along with a tiny decoder-only transformer to produce the desired vocabulary via autoregression. In the next, we scale up the vanilla vision vocabulary by merging the new one with the original one (CLIP), enabling the LVLMs can quickly garner new features. Compared to the popular BLIP-2, MiniGPT4, and LLaVA, Vary can maintain its vanilla capabilities while enjoying more excellent fine-grained perception and understanding ability. Specifically, Vary is competent in new document parsing features (OCR or markdown conversion) while achieving 78.2% ANLS in DocVQA and 36.2% in MMVet. Our code will be publicly available on the homepage.
In this work, we focus on label efficient learning for video action detection. We develop a novel semi-supervised active learning approach which utilizes both labeled as well as unlabeled data along with informative sample selection for action detection. Video action detection requires spatio-temporal localization along with classification, which poses several challenges for both active learning informative sample selection as well as semi-supervised learning pseudo label generation. First, we propose NoiseAug, a simple augmentation strategy which effectively selects informative samples for video action detection. Next, we propose fft-attention, a novel technique based on high-pass filtering which enables effective utilization of pseudo label for SSL in video action detection by emphasizing on relevant activity region within a video. We evaluate the proposed approach on three different benchmark datasets, UCF-101-24, JHMDB-21, and Youtube-VOS. First, we demonstrate its effectiveness on video action detection where the proposed approach outperforms prior works in semi-supervised and weakly-supervised learning along with several baseline approaches in both UCF101-24 and JHMDB-21. Next, we also show its effectiveness on Youtube-VOS for video object segmentation demonstrating its generalization capability for other dense prediction tasks in videos. The code and models is publicly available at: \url{https://github.com/AKASH2907/semi-sup-active-learning}.
3D synthetic-to-real unsupervised domain adaptive segmentation is crucial to annotating new domains. Self-training is a competitive approach for this task, but its performance is limited by different sensor sampling patterns (i.e., variations in point density) and incomplete training strategies. In this work, we propose a density-guided translator (DGT), which translates point density between domains, and integrates it into a two-stage self-training pipeline named DGT-ST. First, in contrast to existing works that simultaneously conduct data generation and feature/output alignment within unstable adversarial training, we employ the non-learnable DGT to bridge the domain gap at the input level. Second, to provide a well-initialized model for self-training, we propose a category-level adversarial network in stage one that utilizes the prototype to prevent negative transfer. Finally, by leveraging the designs above, a domain-mixed self-training method with source-aware consistency loss is proposed in stage two to narrow the domain gap further. Experiments on two synthetic-to-real segmentation tasks (SynLiDAR $\rightarrow$ semanticKITTI and SynLiDAR $\rightarrow$ semanticPOSS) demonstrate that DGT-ST outperforms state-of-the-art methods, achieving 9.4$\%$ and 4.3$\%$ mIoU improvements, respectively. Code is available at \url{https://github.com/yuan-zm/DGT-ST}.
In the realm of 3D computer vision, parametric models have emerged as a ground-breaking methodology for the creation of realistic and expressive 3D avatars. Traditionally, they rely on Principal Component Analysis (PCA), given its ability to decompose data to an orthonormal space that maximally captures shape variations. However, due to the orthogonality constraints and the global nature of PCA's decomposition, these models struggle to perform localized and disentangled editing of 3D shapes, which severely affects their use in applications requiring fine control such as face sculpting. In this paper, we leverage diffusion models to enable diverse and fully localized edits on 3D meshes, while completely preserving the un-edited regions. We propose an effective diffusion masking training strategy that, by design, facilitates localized manipulation of any shape region, without being limited to predefined regions or to sparse sets of predefined control vertices. Following our framework, a user can explicitly set their manipulation region of choice and define an arbitrary set of vertices as handles to edit a 3D mesh. Compared to the current state-of-the-art our method leads to more interpretable shape manipulations than methods relying on latent code state, greater localization and generation diversity while offering faster inference than optimization based approaches. Project page: https://rolpotamias.github.io/Shapefusion/
Large language models (LLMs) are prone to hallucinations, which sparked a widespread effort to detect and prevent them. Recent work attempts to mitigate hallucinations by intervening in the model's generation, typically computing representative vectors of hallucinations vs. grounded generations, for steering the model's hidden states away from a hallucinatory state. However, common studies employ different setups and do not properly separate different possible causes of hallucinations, making interventions misguided. In this work, we introduce a method for categorizing examples based on the model's prior knowledge, named WACK. We construct WACK benchmarks that support interventions in two settings: open-book and closed-book question answering. Using the benchmarks, we perform an extensive investigation of the effect of different choices for intervention, such as the intervened components, and how often and how strongly to intervene. We find that intervention success varies depending on the component, with the attention blocks performing well and the residual stream proving detrimental to language modeling capabilities. We also show that interventions can benefit from representative vectors collected before, rather than after, a hallucination occurs. Finally, we introduce a new dynamic intervention, which intervenes only if needed, and thus is more robust than standard static interventions. The code is available at https://github.com/technion-cs-nlp/hallucination-mitigation .
Despite the success of Transformers on language understanding, code generation, and logical reasoning, they still fail to generalize over length on basic arithmetic tasks such as addition and multiplication. A major reason behind this failure is the vast difference in structure between numbers and text; For example, the numbers are typically parsed from right to left, and there is a correspondence between digits at the same position across different numbers. In contrast, for text, such symmetries are quite unnatural. In this work, we propose to encode these semantics explicitly into the model via modified number formatting and custom positional encodings. Empirically, our method allows a Transformer trained on numbers with at most 5-digits for addition and multiplication to generalize up to 50-digit numbers, without using additional data for longer sequences. We further demonstrate that traditional absolute positional encodings (APE) fail to generalize to longer sequences, even when trained with augmented data that captures task symmetries. To elucidate the importance of explicitly encoding structure, we prove that explicit incorporation of structure via positional encodings is necessary for out-of-distribution generalization. Finally, we pinpoint other challenges inherent to length generalization beyond capturing symmetries, in particular complexity of the underlying task, and propose changes in the training distribution to address them.
Large language models (LLMs) and prompt engineering hold significant potential for advancing computer programming education through personalized instruction. This paper explores this potential by investigating three critical research questions: the systematic categorization of prompt engineering strategies tailored to diverse educational needs, the empowerment of LLMs to solve complex problems beyond their inherent capabilities, and the establishment of a robust framework for evaluating and implementing these strategies. Our methodology involves categorizing programming questions based on educational requirements, applying various prompt engineering strategies, and assessing the effectiveness of LLM-generated responses. Experiments with GPT-4, GPT-4o, Llama3-8b, and Mixtral-8x7b models on datasets such as LeetCode and USACO reveal that GPT-4o consistently outperforms others, particularly with the "multi-step" prompt strategy. The results show that tailored prompt strategies significantly enhance LLM performance, with specific strategies recommended for foundational learning, competition preparation, and advanced problem-solving. This study underscores the crucial role of prompt engineering in maximizing the educational benefits of LLMs. By systematically categorizing and testing these strategies, we provide a comprehensive framework for both educators and students to optimize LLM-based learning experiences. Future research should focus on refining these strategies and addressing current LLM limitations to further enhance educational outcomes in computer programming instruction.