Localization in a battlefield environment is increasingly challenging as GPS connectivity is often denied or unreliable, and physical deployment of anchor nodes across wireless networks for localization can be difficult in hostile battlefield terrain. Existing range-free localization methods rely on radio-based anchors and their average hop distance which suffers from accuracy and stability in dynamic and sparse wireless network topology. Vision-based methods like SLAM and Visual Odometry use expensive sensor fusion techniques for map generation and pose estimation. This paper proposes a novel framework for localization in non-GPS battlefield environments using only the passive camera sensors and considering naturally existing or artificial landmarks as anchors. The proposed method utilizes a customcalibrated stereo vision camera for distance estimation and the YOLOv8s model, which is trained and fine-tuned with our real-world dataset for landmark recognition. The depth images are generated using an efficient stereomatching algorithm, and distances to landmarks are determined by extracting the landmark depth feature utilizing a bounding box predicted by the landmark recognition model. The position of the unknown node is then obtained using the efficient least square algorithm and then optimized using the L-BFGS-B (limited-memory quasi-Newton code for bound-constrained optimization) method. Experimental results demonstrate that our proposed framework performs better than existing anchorbased DV-Hop algorithms and competes with the most efficient vision-based algorithms in terms of localization error (RMSE).
Fine-tuning on generalized tasks such as instruction following, code generation, and mathematics has been shown to enhance language models' performance on a range of tasks. Nevertheless, explanations of how such fine-tuning influences the internal computations in these models remain elusive. We study how fine-tuning affects the internal mechanisms implemented in language models. As a case study, we explore the property of entity tracking, a crucial facet of language comprehension, where models fine-tuned on mathematics have substantial performance gains. We identify the mechanism that enables entity tracking and show that (i) in both the original model and its fine-tuned versions primarily the same circuit implements entity tracking. In fact, the entity tracking circuit of the original model on the fine-tuned versions performs better than the full original model. (ii) The circuits of all the models implement roughly the same functionality: Entity tracking is performed by tracking the position of the correct entity in both the original model and its fine-tuned versions. (iii) Performance boost in the fine-tuned models is primarily attributed to its improved ability to handle the augmented positional information. To uncover these findings, we employ: Patch Patching, DCM, which automatically detects model components responsible for specific semantics, and CMAP, a new approach for patching activations across models to reveal improved mechanisms. Our findings suggest that fine-tuning enhances, rather than fundamentally alters, the mechanistic operation of the model.
Continual learning requires a model to adapt to ongoing changes in the data distribution, and often to the set of tasks to be performed. It is rare, however, that the data and task changes are completely unpredictable. Given a description of an overarching goal or data theme, which we call a realm, humans can often guess what concepts are associated with it. We show here that the combination of a large language model and an image generation model can similarly provide useful premonitions as to how a continual learning challenge might develop over time. We use the large language model to generate text descriptions of semantically related classes that might potentially appear in the data stream in future. These descriptions are then rendered using Stable Diffusion to generate new labelled image samples. The resulting synthetic dataset is employed for supervised pre-training, but is discarded prior to commencing continual learning, along with the pre-training classification head. We find that the backbone of our pre-trained networks can learn representations useful for the downstream continual learning problem, thus becoming a valuable input to any existing continual learning method. Although there are complexities arising from the domain gap between real and synthetic images, we show that pre-training models in this manner improves multiple Class Incremenal Learning (CIL) methods on fine-grained image classification benchmarks. Supporting code can be found at https://github.com/cl-premonition/premonition.
To enhance the controllability of text-to-image diffusion models, existing efforts like ControlNet incorporated image-based conditional controls. In this paper, we reveal that existing methods still face significant challenges in generating images that align with the image conditional controls. To this end, we propose ControlNet++, a novel approach that improves controllable generation by explicitly optimizing pixel-level cycle consistency between generated images and conditional controls. Specifically, for an input conditional control, we use a pre-trained discriminative reward model to extract the corresponding condition of the generated images, and then optimize the consistency loss between the input conditional control and extracted condition. A straightforward implementation would be generating images from random noises and then calculating the consistency loss, but such an approach requires storing gradients for multiple sampling timesteps, leading to considerable time and memory costs. To address this, we introduce an efficient reward strategy that deliberately disturbs the input images by adding noise, and then uses the single-step denoised images for reward fine-tuning. This avoids the extensive costs associated with image sampling, allowing for more efficient reward fine-tuning. Extensive experiments show that ControlNet++ significantly improves controllability under various conditional controls. For example, it achieves improvements over ControlNet by 11.1% mIoU, 13.4% SSIM, and 7.6% RMSE, respectively, for segmentation mask, line-art edge, and depth conditions. All the code, models, demo and organized data have been open sourced on our Github Repo.
We present Splat-MOVER, a modular robotics stack for open-vocabulary robotic manipulation, which leverages the editability of Gaussian Splatting (GSplat) scene representations to enable multi-stage manipulation tasks. Splat-MOVER consists of: (i) ASK-Splat, a GSplat representation that distills semantic and grasp affordance features into the 3D scene. ASK-Splat enables geometric, semantic, and affordance understanding of 3D scenes, which is critical in many robotics tasks; (ii) SEE-Splat, a real-time scene-editing module using 3D semantic masking and infilling to visualize the motions of objects that result from robot interactions in the real-world. SEE-Splat creates a "digital twin" of the evolving environment throughout the manipulation task; and (iii) Grasp-Splat, a grasp generation module that uses ASK-Splat and SEE-Splat to propose affordance-aligned candidate grasps for open-world objects. ASK-Splat is trained in real-time from RGB images in a brief scanning phase prior to operation, while SEE-Splat and Grasp-Splat run in real-time during operation. We demonstrate the superior performance of Splat-MOVER in hardware experiments on a Kinova robot compared to two recent baselines in four single-stage, open-vocabulary manipulation tasks and in four multi-stage manipulation tasks, using the edited scene to reflect changes due to prior manipulation stages, which is not possible with existing baselines. Video demonstrations and the code for the project are available at https://splatmover.github.io.
We introduce the LiLiuM series of large language models (LLMs): 1B, 7B, and 13B parameter models developed 100% in-house to fit eBay's specific needs in the e-commerce domain. This gives eBay full control over all aspects of the models including license, data, vocabulary, and architecture. We expect these models to be used as a foundation for fine-tuning and instruction-tuning, eliminating dependencies to external models. The LiLiuM LLMs have been trained on 3 trillion tokens of multilingual text from general and e-commerce domain. They perform similar to the popular LLaMA-2 models on English natural language understanding (NLU) benchmarks. At the same time, we outperform LLaMA-2 on non-English NLU tasks, machine translation and on e-commerce specific downstream tasks. As part of our data mixture, we utilize the newly released RedPajama-V2 dataset for training and share our insights regarding data filtering and deduplication. We also discuss in detail how to serialize structured data for use in autoregressive language modeling. We provide insights on the effects of including code and parallel machine translation data in pre-training. Furthermore, we develop our own tokenizer and model vocabulary, customized towards e-commerce. This way, we can achieve up to 34% speed-up in text generation on eBay-specific downstream tasks compared to LLaMA-2. Finally, in relation to LLM pretraining, we show that checkpoint averaging can further improve over the best individual model checkpoint.
Autonomous agents powered by language models (LMs) have demonstrated promise in their ability to perform decision-making tasks such as web automation. However, a key limitation remains: LMs, primarily optimized for natural language understanding and generation, struggle with multi-step reasoning, planning, and using environmental feedback when attempting to solve realistic computer tasks. Towards addressing this, we propose an inference-time search algorithm for LM agents to explicitly perform exploration and multi-step planning in interactive web environments. Our approach is a form of best-first tree search that operates within the actual environment space, and is complementary with most existing state-of-the-art agents. It is the first tree search algorithm for LM agents that shows effectiveness on realistic web tasks. On the challenging VisualWebArena benchmark, applying our search algorithm on top of a GPT-4o agent yields a 39.7% relative increase in success rate compared to the same baseline without search, setting a state-of-the-art success rate of 26.4%. On WebArena, search also yields a 28.0% relative improvement over a baseline agent, setting a competitive success rate of 19.2%. Our experiments highlight the effectiveness of search for web agents, and we demonstrate that performance scales with increased test-time compute. We conduct a thorough analysis of our results to highlight improvements from search, limitations, and promising directions for future work. Our code and models are publicly released at https://jykoh.com/search-agents.
Over the past decade, extensive research efforts have been dedicated to the extraction of information from textual process descriptions. Despite the remarkable progress witnessed in natural language processing (NLP), information extraction within the Business Process Management domain remains predominantly reliant on rule-based systems and machine learning methodologies. Data scarcity has so far prevented the successful application of deep learning techniques. However, the rapid progress in generative large language models (LLMs) makes it possible to solve many NLP tasks with very high quality without the need for extensive data. Therefore, we systematically investigate the potential of LLMs for extracting information from textual process descriptions, targeting the detection of process elements such as activities and actors, and relations between them. Using a heuristic algorithm, we demonstrate the suitability of the extracted information for process model generation. Based on a novel prompting strategy, we show that LLMs are able to outperform state-of-the-art machine learning approaches with absolute performance improvements of up to 8\% $F_1$ score across three different datasets. We evaluate our prompting strategy on eight different LLMs, showing it is universally applicable, while also analyzing the impact of certain prompt parts on extraction quality. The number of example texts, the specificity of definitions, and the rigour of format instructions are identified as key for improving the accuracy of extracted information. Our code, prompts, and data are publicly available.
Advancements in Multimodal Large Language Models (MLLMs) have significantly improved medical task performance, such as Visual Question Answering (VQA) and Report Generation (RG). However, the fairness of these models across diverse demographic groups remains underexplored, despite its importance in healthcare. This oversight is partly due to the lack of demographic diversity in existing medical multimodal datasets, which complicates the evaluation of fairness. In response, we propose FMBench, the first benchmark designed to evaluate the fairness of MLLMs performance across diverse demographic attributes. FMBench has the following key features: 1: It includes four demographic attributes: race, ethnicity, language, and gender, across two tasks, VQA and RG, under zero-shot settings. 2: Our VQA task is free-form, enhancing real-world applicability and mitigating the biases associated with predefined choices. 3: We utilize both lexical metrics and LLM-based metrics, aligned with clinical evaluations, to assess models not only for linguistic accuracy but also from a clinical perspective. Furthermore, we introduce a new metric, Fairness-Aware Performance (FAP), to evaluate how fairly MLLMs perform across various demographic attributes. We thoroughly evaluate the performance and fairness of eight state-of-the-art open-source MLLMs, including both general and medical MLLMs, ranging from 7B to 26B parameters on the proposed benchmark. We aim for FMBench to assist the research community in refining model evaluation and driving future advancements in the field. All data and code will be released upon acceptance.
Lately, the practice of utilizing task-specific fine-tuning has been implemented to improve the performance of large language models (LLM) in subsequent tasks. Through the integration of diverse LLMs, the overall competency of LLMs is significantly boosted. Nevertheless, traditional ensemble methods are notably memory-intensive, necessitating the simultaneous loading of all specialized models into GPU memory. To address the inefficiency, model merging strategies have emerged, merging all LLMs into one model to reduce the memory footprint during inference. Despite these advances, model merging often leads to parameter conflicts and performance decline as the number of experts increases. Previous methods to mitigate these conflicts include post-pruning and partial merging. However, both approaches have limitations, particularly in terms of performance and storage efficiency when merged experts increase. To address these challenges, we introduce Channel Merging, a novel strategy designed to minimize parameter conflicts while enhancing storage efficiency. This method clusters and merges channel parameters based on their similarity to form several groups offline. By ensuring that only highly similar parameters are merged within each group, it significantly reduces parameter conflicts. During inference, we can instantly look up the expert parameters from the merged groups, preserving specialized knowledge. Our experiments demonstrate that Channel Merging consistently delivers high performance, matching unmerged models in tasks like English and Chinese reasoning, mathematical reasoning, and code generation. Moreover, it obtains results comparable to model ensemble with just 53% parameters when used with a task-specific router.
Hematoxylin and eosin (H&E) staining is a gold standard for microscopic diagnosis in pathology. However, H&E staining does not capture all the diagnostic information that may be needed. To obtain additional molecular information, immunohistochemical (IHC) stains highlight proteins that mark specific cell types, such as CD3 for T-cells or CK8/18 for epithelial cells. While IHC stains are vital for prognosis and treatment guidance, they are typically only available at specialized centers and time consuming to acquire, leading to treatment delays for patients. Virtual staining, enabled by deep learning-based image translation models, provides a promising alternative by computationally generating IHC stains from H&E stained images. Although many GAN and diffusion based image to image (I2I) translation methods have been used for virtual staining, these models treat image patches as independent data points, which results in increased and more diverse data requirements for effective generation. We present ImplicitStainer, a novel approach that leverages local implicit functions to improve image translation, specifically virtual staining performance, by focusing on pixel-level predictions. This method enhances robustness to variations in dataset sizes, delivering high-quality results even with limited data. We validate our approach on two datasets using a comprehensive set of metrics and benchmark it against over fifteen state-of-the-art GAN- and diffusion based models. Full Code and models trained will be released publicly via Github upon acceptance.
Recently, Graph Neural Networks (GNNs) have become the dominant approach for Knowledge Graph-aware Recommender Systems (KGRSs) due to their proven effectiveness. Building upon GNN-based KGRSs, Self-Supervised Learning (SSL) has been incorporated to address the sparity issue, leading to longer training time. However, through extensive experiments, we reveal that: (1)compared to other KGRSs, the existing GNN-based KGRSs fail to keep their superior performance under sparse interactions even with SSL. (2) More complex models tend to perform worse in sparse interaction scenarios and complex mechanisms, like attention mechanism, can be detrimental as they often increase learning difficulty. Inspired by these findings, we propose LightKG, a simple yet powerful GNN-based KGRS to address sparsity issues. LightKG includes a simplified GNN layer that encodes directed relations as scalar pairs rather than dense embeddings and employs a linear aggregation framework, greatly reducing the complexity of GNNs. Additionally, LightKG incorporates an efficient contrastive layer to implement SSL. It directly minimizes the node similarity in original graph, avoiding the time-consuming subgraph generation and comparison required in previous SSL methods. Experiments on four benchmark datasets show that LightKG outperforms 12 competitive KGRSs in both sparse and dense scenarios while significantly reducing training time. Specifically, it surpasses the best baselines by an average of 5.8\% in recommendation accuracy and saves 84.3\% of training time compared to KGRSs with SSL. Our code is available at https://github.com/1371149/LightKG.
Looped Transformers improve parameter efficiency by applying shared Transformer blocks over recurrent loops, increasing computational depth without increasing the parameter count. However, KV cache memory still scales with loop count, becoming a key memory bottleneck that limits batch size and inference throughput. KV cache quantization can alleviate this bottleneck, but existing methods often suffer substantial accuracy degradation at aggressive low-precision regimes. We observe that looped Transformers offer a unique opportunity: KV states across loops are highly similar. Based on this observation, we propose ResidualQuant, which uses the final-loop KV states as a reference and represents the remaining loops with low-precision residuals. Our method further combines least-square scaling and rotation-based quantization, as well as loop-wise mixed precision, to enable accurate quantization down to INT2 while retaining efficient reconstruction. Across multiple looped Transformer models and mathematical reasoning and code generation benchmarks, ResidualQuant consistently improves the accuracy-memory tradeoff over state-of-the-art rotation-based KV quantization. In particular, our method retains accuracy close to BF16 under mixed-precision settings while reducing theoretical KV storage by 80.7%, achieving up to 13.0% higher accuracy than the rotation-based baseline at nearly the same memory budget. On an RTX 5090, reduced KV memory traffic improves fixed-batch decode throughput by up to 3.66x, while the smaller memory footprint enables up to 8x larger batches, improving peak throughput by up to 5.38x.
모달리티 간 강건한 의료 이미지 분할은 심각한 도메인 편이와 타겟 도메인 레이블의 부재로 인해 여전히 어려운 과제입니다. 확산 모델이 교차 도메인 생성 및 증강에 활용되어 왔지만, 타겟 도메인 조건부 확산 학습은 일반적으로 노이즈가 심한 의사 마스크에 의존하며, 단일 Arg-Max 의사 레이블에 단순히 조건을 부여하면 확산 학습 및 하위 이미지 분할을 손상시킬 수 있습니다. 우리는 타겟 도메인 조건부 확산을 위한 불확실성 기반 학습 목표를 핵심으로 하는 통합 비지도 도메인 적응(UDA) 프레임워크인 UPDiff-UDA를 제안합니다. 불완전한 소스 학습 분할기가 주어지면, 픽셀별 소프트맥스 분포를 활용하여 순위가 매겨진 의사 레이블 맵(Arg-Max, Arg-2nd, Arg-3rd, ...)을 구성합니다. 각 맵은 조건부 스코어 추정값을 산출하며, 이를 픽셀 단위 신뢰도 가중치로 집계하여 스코어 매칭을 위한 불확실성 재가중 스코어를 얻습니다. 이를 통해 불확실한 영역에서 대안적인 그럴듯한 레이블을 활용하면서 의사 레이블 노이즈에 대한 강건성을 향상시킵니다. 또한, 신뢰도 가중 집계가 분할기 유도 대리 레이블 분포 하에서 최소-MSE 볼록 결합 원리를 따름을 이론적으로 정당화합니다. 의사 조건의 품질을 향상시키기 위해, 외관 격차를 줄이는 특징 기반 저자유도 베지어(Bézier) 곡선 적응 방법도 도입합니다. 여러 공개 데이터셋과 모달리티 전환에 대한 실험을 통해, UPDiff-UDA가 증강을 위한 고충실도 레이블링된 타겟 스타일 샘플을 생성하고 강력한 UDA 기준선들을 일관되게 능가함을 보입니다. 프로젝트 코드는 https://github.com/superlc1995/Multi-Channel-Uncertainty-Diffusion-UDA 에서 확인할 수 있습니다.
고대 한자에 대한 학술적 훈고(주석)는 시각적 관찰, 언어학적 분석, 역사적 맥락의 통합을 요구합니다. 그러나 기존의 전산적 접근 방식은 문자 인식 및 검색과 같은 하위 작업에만 좁게 집중되어 있어, 포괄적인 학술 분석에 필요한 정형화된 데이터셋과 벤치마크가 부족합니다. 이러한 한계를 해결하기 위해, 본 논문에서는 학술적 훈고 과정을 모델링한 비전-언어 시각적 질문 답변(VQA) 태스크인 고대 한자 주석(ACCE)을 도입합니다. ACCE는 기본 문자 식별, 자형 분석, 의미 주석, 통시적 진화 분석의 4단계 점진적 수준으로 구성됩니다. 이 태스크를 지원하기 위해 상호 보완적인 두 가지 자원을 구축합니다. JieZi-Dataset은 ACCE를 위해 구축된 최초의 대규모 전문가 검증 VQA 학습 데이터셋으로, 50만 개 이상의 QA 쌍으로 구성되어 있습니다. 이는 전문가가 설계한 템플릿과 원전 참조를 통해 생성을 제한함으로써 사실적 오류를 줄이는 파이프라인을 거쳐 구축됩니다. 학술적 정확성을 보장하기 위해 각 핵심 단계에서 인간 검증이 추가로 적용됩니다. JieZi-Bench는 훈고 과정에 부합하도록 전문가에 의해 구축 및 검증되어 평가 신뢰성을 보장하는 평가 벤치마크입니다. 이는 학습 데이터와 엄격히 분리된 권위 있는 사전 편찬 저작물에서 선별한 참조 답변을 갖춘 4단계로 구성됩니다. 멀티모달 대규모 언어 모델(MLLM)을 대상으로 한 실험에 따르면, 현재 모델들은 기본적인 식별 작업에서는 우수한 성능을 보이지만 자형 분석, 의미 추론, 통시적 이해에서는 어려움을 겪는 것으로 나타났습니다. JieZi-Dataset을 활용한 미세조정은 4개 수준 전반에서 성능을 크게 향상시킵니다. 코드와 데이터셋은 https://github.com/Ran00w/JieZi 에서 확인할 수 있습니다.
대규모 언어 모델(LLM)은 단순 코드 완성을 넘어 컨텍스트를 검색하고, 파일을 편집하며, 도구를 실행하고, 보안에 민감한 워크플로에 참여하는 저장소 규모의 에이전트로 발전하고 있다. 그러나 이러한 시스템에 대한 실증적 증거는 기능적 작업 완료 중심의 소프트웨어 공학 평가와 취약점 탐지, 보안 생성, 또는 익스플로잇 기반 검증 중심의 소프트웨어 보안 평가로 양분되어 있다. 본 증거 중심 구조적 서베이는 2026년 5월 31일까지 발표된 대표 연구들을 소프트웨어 공학 작업, 소프트웨어 보안 작업, 적응 메커니즘, 산출물 세분성, 평가 설계 전반에 걸쳐 종합한다. 작업 분류 체계와 더불어, 기능적 정확성, 보안성, 운영 신뢰성, 증거 출처, 에이전트 권한을 분리하는 보증 프레임워크를 소개한다. 검토 결과, 실행 피드백과 저장소 접근은 엔지니어링 작업 완료율을 크게 향상시킬 수 있지만 그 자체로 보안성을 확립하지는 못하며, 반대로 정적 분석 라벨이나 취약점 분류 점수가 실제 배포 가능한 수준의 정확성을 보장하는 경우는 드물다는 점을 확인했다. 또한 취약한 테스트 오라클, 중복 및 시간적으로 유출된 데이터, 변동되는 에이전트 하네스, 프록시 전용 보안 점검, 불충분하게 보고된 예산 및 사람의 개입 등 반복되는 타당성 위협을 식별하고, 연구 간 비교를 위한 최소 보고 프로토콜을 도출했다. 이를 바탕으로 도출된 연구 과제는 보안성과 기능성을 동시에 다루는 벤치마크, 저장소 규모의 위협 모델, 정밀하게 보정된 인간 감독, 장기적 유지보수성 증거, 재현 가능한 에이전트 평가를 우선순위로 둔다. 핵심 결론은 모델의 능력이 단일 벤치마크 점수가 아니라 작업에 적합한 증거로 뒷받침되는 보증 사례(assurance case)로 평가되어야 한다는 점이다.
파노라마 카메라는 360° 시야각을 제공하여 파노라마 비전, 몰입형 시각 컴퓨팅, 로봇 인식 분야에서 널리 활용되고 있다. 그러나 카메라 방향의 변화는 수직이 맞지 않는(non-upright) 파노라마를 생성하여 기하학적 변형을 유발하며, 이는 다운스트림 시각 분석을 복잡하게 만들 수 있다. 기존의 비전 기반 보정 방법들은 대개 단일 투영 도메인 내에서 작동하므로, 국소적 기하학적 구조와 전역적 문맥 정보를 함께 활용하는 능력에 한계가 있다. 이를 해결하기 위해 본 논문에서는 360° 이미지 왜곡 보정을 투영 인식(projection-aware) 표현 학습 문제로 정의하고, 직립 파노라마 보정을 위한 이중 투영 프레임워크를 제안한다. 합성곱 신경망(CNN) 브랜치는 정거원통도법(ERP) 입력으로부터 국소적 기하 구조를 포착하며, 비전 트랜스포머(ViT) 브랜치는 큐브맵 투영으로부터 전역적 문맥 단서를 모델링한다. 상호 투영 특징 변환 및 다단계 특징 융합을 통해 이러한 상호 보완적인 표현 간의 효과적인 상호작용이 가능해진다. 학습된 표현은 협력적 경사 추정과 직립 파노라마 생성을 지원하며, 이 두 작업은 상호 보완적인 기하학적 및 외관적 지도(supervision)를 제공한다. SUN360 및 M3D 데이터셋 대상 실험에서 본 기법은 기존 방법 대비 일관된 성능 향상을 보여주었으며, 각각 1° 오차 임계값 내 정확도 65.9% 및 85.2%, Fréchet Inception Distance 점수 5.87 및 3.26을 달성하였다. 소거 연구(ablation studies)를 통해 이중 투영 표현, 상호 투영 특징 변환 및 융합, 협력적 다중 작업 학습의 기여도를 검증하였다. 제안하는 프레임워크는 파노라마 보정을 위한 투영 인식 시각 컴퓨팅 접근법을 제공한다. 코드, 사전학습 모델, 학습/테스트 스크립트는 https://github.com/YuhaoShine/DualProjectionFusion 에서 확인할 수 있다.
Organ segmentation in CT volumes is an important pre-processing step in many computer assisted intervention and diagnosis methods. In recent years, convolutional neural networks have dominated the state of the art in this task. However, since this problem presents a challenging environment due to high variability in the organ's shape and similarity between tissues, the generation of false negative and false positive regions in the output segmentation is a common issue. Recent works have shown that the uncertainty analysis of the model can provide us with useful information about potential errors in the segmentation. In this context, we proposed a segmentation refinement method based on uncertainty analysis and graph convolutional networks. We employ the uncertainty levels of the convolutional network in a particular input volume to formulate a semi-supervised graph learning problem that is solved by training a graph convolutional network. To test our method we refine the initial output of a 2D U-Net. We validate our framework with the NIH pancreas dataset and the spleen dataset of the medical segmentation decathlon. We show that our method outperforms the state-of-the art CRF refinement method by improving the dice score by 1% for the pancreas and 2% for spleen, with respect to the original U-Net's prediction. Finally, we discuss the results and current limitations of the model for future work in this research direction. For reproducibility purposes, we make our code publicly available.
Weakly supervised semantic segmentation (WSSS) with image-level labels is a challenging task. Mainstream approaches follow a multi-stage framework and suffer from high training costs. In this paper, we explore the potential of Contrastive Language-Image Pre-training models (CLIP) to localize different categories with only image-level labels and without further training. To efficiently generate high-quality segmentation masks from CLIP, we propose a novel WSSS framework called CLIP-ES. Our framework improves all three stages of WSSS with special designs for CLIP: 1) We introduce the softmax function into GradCAM and exploit the zero-shot ability of CLIP to suppress the confusion caused by non-target classes and backgrounds. Meanwhile, to take full advantage of CLIP, we re-explore text inputs under the WSSS setting and customize two text-driven strategies: sharpness-based prompt selection and synonym fusion. 2) To simplify the stage of CAM refinement, we propose a real-time class-aware attention-based affinity (CAA) module based on the inherent multi-head self-attention (MHSA) in CLIP-ViTs. 3) When training the final segmentation model with the masks generated by CLIP, we introduced a confidence-guided loss (CGL) focus on confident regions. Our CLIP-ES achieves SOTA performance on Pascal VOC 2012 and MS COCO 2014 while only taking 10% time of previous methods for the pseudo mask generation. Code is available at https://github.com/linyq2117/CLIP-ES.
Panoramic radiography (Panoramic X-ray, PX) is a widely used imaging modality for dental examination. However, PX only provides a flattened 2D image, lacking in a 3D view of the oral structure. In this paper, we propose NeBLa (Neural Beer-Lambert) to estimate 3D oral structures from real-world PX. NeBLa tackles full 3D reconstruction for varying subjects (patients) where each reconstruction is based only on a single panoramic image. We create an intermediate representation called simulated PX (SimPX) from 3D Cone-beam computed tomography (CBCT) data based on the Beer-Lambert law of X-ray rendering and rotational principles of PX imaging. SimPX aims at not only truthfully simulating PX, but also facilitates the reverting process back to 3D data. We propose a novel neural model based on ray tracing which exploits both global and local input features to convert SimPX to 3D output. At inference, a real PX image is translated to a SimPX-style image with semantic regularization, and the translated image is processed by generation module to produce high-quality outputs. Experiments show that NeBLa outperforms prior state-of-the-art in reconstruction tasks both quantitatively and qualitatively. Unlike prior methods, NeBLa does not require any prior information such as the shape of dental arches, nor the matched PX-CBCT dataset for training, which is difficult to obtain in clinical practice. Our code is available at https://github.com/sihwa-park/nebla.