Abstract:We address the problem of video question answering (video QA) with temporal grounding in a weakly supervised setup, without any temporal annotations. Given a video and a question, we generate an open-ended answer grounded with the start and end time. For this task, we propose TOGA: a vision-language model for Temporally Grounded Open-Ended Video QA with Weak Supervision. We instruct-tune TOGA to jointly generate the answer and the temporal grounding. We operate in a weakly supervised setup where the temporal grounding annotations are not available. We generate pseudo labels for temporal grounding and ensure the validity of these labels by imposing a consistency constraint between the question of a grounding response and the response generated by a question referring to the same temporal segment. We notice that jointly generating the answers with the grounding improves performance on question answering as well as grounding. We evaluate TOGA on grounded QA and open-ended QA tasks. For grounded QA, we consider the NExT-GQA benchmark which is designed to evaluate weakly supervised grounded question answering. For open-ended QA, we consider the MSVD-QA and ActivityNet-QA benchmarks. We achieve state-of-the-art performance for both tasks on these benchmarks.
Abstract:As a cornerstone of patient care, clinical decision-making significantly influences patient outcomes and can be enhanced by large language models (LLMs). Although LLMs have demonstrated remarkable performance, their application to visual question answering in medical imaging, particularly for reasoning-based diagnosis, remains largely unexplored. Furthermore, supervised fine-tuning for reasoning tasks is largely impractical due to limited data availability and high annotation costs. In this work, we introduce a zero-shot framework for reliable medical image diagnosis that enhances the reasoning capabilities of LLMs in clinical settings through test-time scaling. Given a medical image and a textual prompt, a vision-language model processes a medical image along with a corresponding textual prompt to generate multiple descriptions or interpretations of visual features. These interpretations are then fed to an LLM, where a test-time scaling strategy consolidates multiple candidate outputs into a reliable final diagnosis. We evaluate our approach across various medical imaging modalities -- including radiology, ophthalmology, and histopathology -- and demonstrate that the proposed test-time scaling strategy enhances diagnostic accuracy for both our and baseline methods. Additionally, we provide an empirical analysis showing that the proposed approach, which allows unbiased prompting in the first stage, improves the reliability of LLM-generated diagnoses and enhances classification accuracy.
Abstract:Vision Transformers (ViTs) have demonstrated impressive performance across a wide range of biometric tasks, including face and body recognition. In this work, we adapt a ViT model pretrained on visible (VIS) imagery to the challenging problem of cross-spectral body recognition, which involves matching images captured in the visible and infrared (IR) domains. Recent ViT architectures have explored incorporating additional embeddings beyond traditional positional embeddings. Building on this idea, we integrate Side Information Embedding (SIE) and examine the impact of encoding domain and camera information to enhance cross-spectral matching. Surprisingly, our results show that encoding only camera information - without explicitly incorporating domain information - achieves state-of-the-art performance on the LLCM dataset. While occlusion handling has been extensively studied in visible-spectrum person re-identification (Re-ID), occlusions in visible-infrared (VI) Re-ID remain largely underexplored - primarily because existing VI-ReID datasets, such as LLCM, SYSU-MM01, and RegDB, predominantly feature full-body, unoccluded images. To address this gap, we analyze the impact of range-induced occlusions using the IARPA Janus Benchmark Multi-Domain Face (IJB-MDF) dataset, which provides a diverse set of visible and infrared images captured at various distances, enabling cross-range, cross-spectral evaluations.
Abstract:Low-Rank Adaptation (LoRA) has gained prominence as a computationally efficient method for fine-tuning generative models, enabling distinct visual concept synthesis with minimal overhead. However, current methods struggle to effectively merge multiple LoRA adapters without training, particularly in complex compositions involving diverse visual elements. We introduce MultLFG, a novel framework for training-free multi-LoRA composition that utilizes frequency-domain guidance to achieve adaptive fusion of multiple LoRAs. Unlike existing methods that uniformly aggregate concept-specific LoRAs, MultLFG employs a timestep and frequency subband adaptive fusion strategy, selectively activating relevant LoRAs based on content relevance at specific timesteps and frequency bands. This frequency-sensitive guidance not only improves spatial coherence but also provides finer control over multi-LoRA composition, leading to more accurate and consistent results. Experimental evaluations on the ComposLoRA benchmark reveal that MultLFG substantially enhances compositional fidelity and image quality across various styles and concept sets, outperforming state-of-the-art baselines in multi-concept generation tasks. Code will be released.
Abstract:Conformal prediction quantifies the uncertainty of machine learning models by augmenting point predictions with valid prediction sets, assuming exchangeability. For complex scenarios involving multiple trials, models, or data sources, conformal prediction sets can be aggregated to create a prediction set that captures the overall uncertainty, often improving precision. However, aggregating multiple prediction sets with individual $1-\alpha$ coverage inevitably weakens the overall guarantee, typically resulting in $1-2\alpha$ worst-case coverage. In this work, we propose a framework for the weighted aggregation of prediction sets, where weights are assigned to each prediction set based on their contribution. Our framework offers flexible control over how the sets are aggregated, achieving tighter coverage bounds that interpolate between the $1-2\alpha$ guarantee of the combined models and the $1-\alpha$ guarantee of an individual model depending on the distribution of weights. We extend our framework to data-dependent weights, and we derive a general procedure for data-dependent weight aggregation that maintains finite-sample validity. We demonstrate the effectiveness of our methods through experiments on synthetic and real data in the mixture-of-experts setting, and we show that aggregation with data-dependent weights provides a form of adaptive coverage.
Abstract:Vision Transformers (ViTs) have delivered remarkable progress through global self-attention, yet their quadratic complexity can become prohibitive for high-resolution inputs. In this work, we present ViT-Linearizer, a cross-architecture distillation framework that transfers rich ViT representations into a linear-time, recurrent-style model. Our approach leverages 1) activation matching, an intermediate constraint that encourages student to align its token-wise dependencies with those produced by the teacher, and 2) masked prediction, a contextual reconstruction objective that requires the student to predict the teacher's representations for unseen (masked) tokens, to effectively distill the quadratic self-attention knowledge into the student while maintaining efficient complexity. Empirically, our method provides notable speedups particularly for high-resolution tasks, significantly addressing the hardware challenges in inference. Additionally, it also elevates Mamba-based architectures' performance on standard vision benchmarks, achieving a competitive 84.3% top-1 accuracy on ImageNet with a base-sized model. Our results underscore the good potential of RNN-based solutions for large-scale visual tasks, bridging the gap between theoretical efficiency and real-world practice.
Abstract:In this work, we tackle the problem of text-to-video retrieval (T2VR). Inspired by the success of late interaction techniques in text-document, text-image, and text-video retrieval, our approach, Video-ColBERT, introduces a simple and efficient mechanism for fine-grained similarity assessment between queries and videos. Video-ColBERT is built upon 3 main components: a fine-grained spatial and temporal token-wise interaction, query and visual expansions, and a dual sigmoid loss during training. We find that this interaction and training paradigm leads to strong individual, yet compatible, representations for encoding video content. These representations lead to increases in performance on common text-to-video retrieval benchmarks compared to other bi-encoder methods.
Abstract:Camera-based monitoring of vital signs, also known as imaging photoplethysmography (iPPG), has seen applications in driver-monitoring, perfusion assessment in surgical settings, affective computing, and more. iPPG involves sensing the underlying cardiac pulse from video of the skin and estimating vital signs such as the heart rate or a full pulse waveform. Some previous iPPG methods impose model-based sparse priors on the pulse signals and use iterative optimization for pulse wave recovery, while others use end-to-end black-box deep learning methods. In contrast, we introduce methods that combine signal processing and deep learning methods in an inverse problem framework. Our methods estimate the underlying pulse signal and heart rate from facial video by learning deep-network-based denoising operators that leverage deep algorithm unfolding and deep equilibrium models. Experiments show that our methods can denoise an acquired signal from the face and infer the correct underlying pulse rate, achieving state-of-the-art heart rate estimation performance on well-known benchmarks, all with less than one-fifth the number of learnable parameters as the closest competing method.
Abstract:Biometric recognition becomes increasingly challenging as we move away from the visible spectrum to infrared imagery, where domain discrepancies significantly impact identification performance. In this paper, we show that body embeddings perform better than face embeddings for cross-spectral person identification in medium-wave infrared (MWIR) and long-wave infrared (LWIR) domains. Due to the lack of multi-domain datasets, previous research on cross-spectral body identification - also known as Visible-Infrared Person Re-Identification (VI-ReID) - has primarily focused on individual infrared bands, such as near-infrared (NIR) or LWIR, separately. We address the multi-domain body recognition problem using the IARPA Janus Benchmark Multi-Domain Face (IJB-MDF) dataset, which enables matching of short-wave infrared (SWIR), MWIR, and LWIR images against RGB (VIS) images. We leverage a vision transformer architecture to establish benchmark results on the IJB-MDF dataset and, through extensive experiments, provide valuable insights into the interrelation of infrared domains, the adaptability of VIS-pretrained models, the role of local semantic features in body-embeddings, and effective training strategies for small datasets. Additionally, we show that finetuning a body model, pretrained exclusively on VIS data, with a simple combination of cross-entropy and triplet losses achieves state-of-the-art mAP scores on the LLCM dataset.
Abstract:Image matching is a key component of modern 3D vision algorithms, essential for accurate scene reconstruction and localization. MASt3R redefines image matching as a 3D task by leveraging DUSt3R and introducing a fast reciprocal matching scheme that accelerates matching by orders of magnitude while preserving theoretical guarantees. This approach has gained strong traction, with DUSt3R and MASt3R collectively cited over 250 times in a short span, underscoring their impact. However, despite its accuracy, MASt3R's inference speed remains a bottleneck. On an A40 GPU, latency per image pair is 198.16 ms, mainly due to computational overhead from the ViT encoder-decoder and Fast Reciprocal Nearest Neighbor (FastNN) matching. To address this, we introduce Speedy MASt3R, a post-training optimization framework that enhances inference efficiency while maintaining accuracy. It integrates multiple optimization techniques, including FlashMatch-an approach leveraging FlashAttention v2 with tiling strategies for improved efficiency, computation graph optimization via layer and tensor fusion having kernel auto-tuning with TensorRT (GraphFusion), and a streamlined FastNN pipeline that reduces memory access time from quadratic to linear while accelerating block-wise correlation scoring through vectorized computation (FastNN-Lite). Additionally, it employs mixed-precision inference with FP16/FP32 hybrid computations (HybridCast), achieving speedup while preserving numerical precision. Evaluated on Aachen Day-Night, InLoc, 7-Scenes, ScanNet1500, and MegaDepth1500, Speedy MASt3R achieves a 54% reduction in inference time (198 ms to 91 ms per image pair) without sacrificing accuracy. This advancement enables real-time 3D understanding, benefiting applications like mixed reality navigation and large-scale 3D scene reconstruction.