-
AdaFocus: Adaptive Relevance-Diversity Sampling with Zero-Cache Look-back for Efficient Long Video Understanding
Authors:
Xiao Yang,
Yingzhe Ma,
Haoxuan Yu,
Zixin Li,
Ning Qin
Abstract:
Long video understanding is heavily bottlenecked by a rigid one-shot paradigm: existing methods either densely encode videos at prohibitive memory and latency costs, or aggressively compress them into sparse frame sets that irreversibly discard fine-grained evidence needed for downstream reasoning. Consequently, current models struggle to simultaneously balance temporal coverage, visual details, a…
▽ More
Long video understanding is heavily bottlenecked by a rigid one-shot paradigm: existing methods either densely encode videos at prohibitive memory and latency costs, or aggressively compress them into sparse frame sets that irreversibly discard fine-grained evidence needed for downstream reasoning. Consequently, current models struggle to simultaneously balance temporal coverage, visual details, and computational efficiency.
We propose AdaFocus, an efficient framework that rethinks long-video understanding as progressive evidence acquisition rather than one-pass encoding. AdaFocus relies on two tightly coupled components. First, a Query-Aware Adaptive Relevance-Diversity sampler (AdaRD) produces a compact yet informative video preview, adaptively switching to global clustering when the query lacks reliable local grounding. Second, instead of caching exhaustive frame sequences in memory, AdaFocus introduces an uncertainty-triggered refinement mechanism. It performs targeted look-back only when the model is not confident, retrieving high-resolution evidence directly from disk via a zero-cache I/O design. This turns discarded visual details from an irreversible loss into on-demand recoverable evidence without paying the cost of exhaustive preloading.
Experiments on seven standard long-video benchmarks show that AdaFocus delivers a substantially better efficiency-accuracy trade-off than strong baselines. Compared with conventional dense encoding, AdaFocus achieves improved task performance (e.g., +2.59 accuracy on VideoMME, +8.39 mIoU on Charades-STA over single-pass inference) while reducing visual token consumption by ~33x and eliminating the need for in-memory frame pre-caching through its zero-cache disk retrieval design. These findings suggest that progressive preview combined with zero-cache evidence refinement is a highly effective paradigm for scalable multimedia reasoning.
△ Less
Submitted 12 May, 2026;
originally announced May 2026.
-
SparseGF: A Height-Aware Sparse Segmentation Framework with Context Compression for Robust Ground Filtering Across Urban to Natural Scenes
Authors:
Nannan Qin,
Pengjie Tao,
Haiyan Guan,
Zhizhong Kang,
Lingfei Ma,
Xiangyun Hu,
Jonathan Li
Abstract:
High-quality digital terrain models derived from airborne laser scanning (ALS) data are essential for a wide range of geospatial analyses, and their generation typically relies on robust ground filtering (GF) to separate point clouds across diverse landscapes into ground and non-ground parts. Although current deep-learning-based GF methods have demonstrated impressive performance, especially in sp…
▽ More
High-quality digital terrain models derived from airborne laser scanning (ALS) data are essential for a wide range of geospatial analyses, and their generation typically relies on robust ground filtering (GF) to separate point clouds across diverse landscapes into ground and non-ground parts. Although current deep-learning-based GF methods have demonstrated impressive performance, especially in specific challenging terrains, their cross-scene generalization remains limited by two persistent issues: the context-detail dilemma in large-scale processing due to limited computational resources, and the random misclassification of tall objects arising from classification-only optimization. To overcome these limitations, we propose SparseGF, a height-aware sparse segmentation framework enhanced with context compression. It is built upon three key innovations: (1) a convex-mirror-inspired context compression module that condenses expansive contexts into compact representations while preserving central details; (2) a hybrid sparse voxel-point network architecture that effectively interprets compressed representations while mitigating compression-induced geometric distortion; and (3) a height-aware loss function that explicitly enforces topographic elevation priors during training to suppress random misclassification of tall objects. Extensive evaluations on two large-scale ALS benchmark datasets demonstrate that SparseGF delivers robust GF across urban to natural terrains, achieving leading performance in complex urban scenes, competitive results on mixed terrains, and moderate yet non-catastrophic accuracy in densely forested steep areas. This work offers new insights into deep-learning-based GF research and encourages further exploration toward truly cross-scene generalization for large-scale environmental monitoring.
△ Less
Submitted 23 April, 2026;
originally announced April 2026.
-
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning
Authors:
NVIDIA,
:,
Aakshita Chandiramani,
Aaron Blakeman,
Abdullahi Olaoye,
Abhibha Gupta,
Abhilash Somasamudramath,
Abhinav Khattar,
Adeola Adesoba,
Adi Renduchintala,
Adil Asif,
Aditya Agrawal,
Aditya Vavre,
Ahmad Kiswani,
Aishwarya Padmakumar,
Ajay Hotchandani,
Akanksha Shukla,
Akhiad Bercovich,
Aleksander Ficek,
Aleksandr Shaposhnikov,
Alex Gronskiy,
Alex Kondratenko,
Alex Neefus,
Alex Steiner,
Alex Yang
, et al. (522 additional authors not shown)
Abstract:
We describe the pre-training, post-training, and quantization of Nemotron 3 Super, a 120 billion (active 12 billion) parameter hybrid Mamba-Attention Mixture-of-Experts model. Nemotron 3 Super is the first model in the Nemotron 3 family to 1) be pre-trained in NVFP4, 2) leverage LatentMoE, a new Mixture-of-Experts architecture that optimizes for both accuracy per FLOP and accuracy per parameter, a…
▽ More
We describe the pre-training, post-training, and quantization of Nemotron 3 Super, a 120 billion (active 12 billion) parameter hybrid Mamba-Attention Mixture-of-Experts model. Nemotron 3 Super is the first model in the Nemotron 3 family to 1) be pre-trained in NVFP4, 2) leverage LatentMoE, a new Mixture-of-Experts architecture that optimizes for both accuracy per FLOP and accuracy per parameter, and 3) include MTP layers for inference acceleration through native speculative decoding. We pre-trained Nemotron 3 Super on 25 trillion tokens followed by post-training using supervised fine tuning (SFT) and reinforcement learning (RL). The final model supports up to 1M context length and achieves comparable accuracy on common benchmarks, while also achieving up to 2.2x and 7.5x higher inference throughput compared to GPT-OSS-120B and Qwen3.5-122B, respectively. Nemotron 3 Super datasets, along with the base, post-trained, and quantized checkpoints, are open-sourced on HuggingFace.
△ Less
Submitted 14 April, 2026;
originally announced April 2026.
-
SOL-ExecBench: Speed-of-Light Benchmarking for Real-World GPU Kernels Against Hardware Limits
Authors:
Edward Lin,
Sahil Modi,
Siva Kumar Sastry Hari,
Qijing Huang,
Zhifan Ye,
Nestor Qin,
Fengzhe Zhou,
Yuan Zhang,
Jingquan Wang,
Sana Damani,
Dheeraj Peri,
Ouye Xie,
Aditya Kane,
Moshe Maor,
Michael Behar,
Triston Cao,
Rishabh Mehta,
Vartika Singh,
Vikram Sharma Mailthody,
Terry Chen,
Zihao Ye,
Hanfeng Chen,
Tianqi Chen,
Vinod Grover,
Wei Chen
, et al. (8 additional authors not shown)
Abstract:
As agentic AI systems become increasingly capable of generating and optimizing GPU kernels, progress is constrained by benchmarks that reward speedup over software baselines rather than proximity to hardware-efficient execution. We present SOL-ExecBench, a benchmark of 235 CUDA kernel optimization problems extracted from 124 production and emerging AI models spanning language, diffusion, vision, a…
▽ More
As agentic AI systems become increasingly capable of generating and optimizing GPU kernels, progress is constrained by benchmarks that reward speedup over software baselines rather than proximity to hardware-efficient execution. We present SOL-ExecBench, a benchmark of 235 CUDA kernel optimization problems extracted from 124 production and emerging AI models spanning language, diffusion, vision, audio, video, and hybrid architectures, targeting NVIDIA Blackwell GPUs. The benchmark covers forward and backward workloads across BF16, FP8, and NVFP4, including kernels whose best performance is expected to rely on Blackwell-specific capabilities. Unlike prior benchmarks that evaluate kernels primarily relative to software implementations, SOL-ExecBench measures performance against analytically derived Speed-of-Light (SOL) bounds computed by SOLAR, our pipeline for deriving hardware-grounded SOL bounds, yielding a fixed target for hardware-efficient optimization. We report a SOL Score that quantifies how much of the gap between a release-defined scoring baseline and the hardware SOL bound a candidate kernel closes. To support robust evaluation of agentic optimizers, we additionally provide a sandboxed harness with GPU clock locking, L2 cache clearing, isolated subprocess execution, and static analysis based checks against common reward-hacking strategies. SOL-ExecBench reframes GPU kernel benchmarking from beating a mutable software baseline to closing the remaining gap to hardware Speed-of-Light.
△ Less
Submitted 19 March, 2026;
originally announced March 2026.
-
MAPI-GNN: Multi-Activation Plane Interaction Graph Neural Network for Multimodal Medical Diagnosis
Authors:
Ziwei Qin,
Xuhui Song,
Deqing Huang,
Na Qin,
Jun Li
Abstract:
Graph neural networks are increasingly applied to multimodal medical diagnosis for their inherent relational modeling capabilities. However, their efficacy is often compromised by the prevailing reliance on a single, static graph built from indiscriminate features, hindering the ability to model patient-specific pathological relationships. To this end, the proposed Multi-Activation Plane Interacti…
▽ More
Graph neural networks are increasingly applied to multimodal medical diagnosis for their inherent relational modeling capabilities. However, their efficacy is often compromised by the prevailing reliance on a single, static graph built from indiscriminate features, hindering the ability to model patient-specific pathological relationships. To this end, the proposed Multi-Activation Plane Interaction Graph Neural Network (MAPI-GNN) reconstructs this single-graph paradigm by learning a multifaceted graph profile from semantically disentangled feature subspaces. The framework first uncovers latent graph-aware patterns via a multi-dimensional discriminator; these patterns then guide the dynamic construction of a stack of activation graphs; and this multifaceted profile is finally aggregated and contextualized by a relational fusion engine for a robust diagnosis. Extensive experiments on two diverse tasks, comprising over 1300 patient samples, demonstrate that MAPI-GNN significantly outperforms state-of-the-art methods.
△ Less
Submitted 22 December, 2025;
originally announced December 2025.
-
Reverb: Open-Source ASR and Diarization from Rev
Authors:
Nishchal Bhandari,
Danny Chen,
Miguel Ángel del Río Fernández,
Natalie Delworth,
Jennifer Drexler Fox,
Migüel Jetté,
Quinten McNamara,
Corey Miller,
Ondřej Novotný,
Ján Profant,
Nan Qin,
Martin Ratajczak,
Jean-Philippe Robichaud
Abstract:
Today, we are open-sourcing our core speech recognition and diarization models for non-commercial use. We are releasing both a full production pipeline for developers as well as pared-down research models for experimentation. Rev hopes that these releases will spur research and innovation in the fast-moving domain of voice technology. The speech recognition models released today outperform all exi…
▽ More
Today, we are open-sourcing our core speech recognition and diarization models for non-commercial use. We are releasing both a full production pipeline for developers as well as pared-down research models for experimentation. Rev hopes that these releases will spur research and innovation in the fast-moving domain of voice technology. The speech recognition models released today outperform all existing open source speech recognition models across a variety of long-form speech recognition domains.
△ Less
Submitted 24 February, 2025; v1 submitted 4 October, 2024;
originally announced October 2024.
-
TIA: A Teaching Intonation Assessment Dataset in Real Teaching Situations
Authors:
Shuhua Liu,
Chunyu Zhang,
Binshuai Li,
Niantong Qin,
Huanting Cheng,
Huayu Zhang
Abstract:
Intonation is one of the important factors affecting the teaching language arts, so it is an urgent problem to be addressed by evaluating the teachers' intonation through artificial intelligence technology. However, the lack of an intonation assessment dataset has hindered the development of the field. To this end, this paper constructs a Teaching Intonation Assessment (TIA) dataset for the first…
▽ More
Intonation is one of the important factors affecting the teaching language arts, so it is an urgent problem to be addressed by evaluating the teachers' intonation through artificial intelligence technology. However, the lack of an intonation assessment dataset has hindered the development of the field. To this end, this paper constructs a Teaching Intonation Assessment (TIA) dataset for the first time in real teaching situations. This dataset covers 9 disciplines, 396 teachers, total of 11,444 utterance samples with a length of 15 seconds. In order to test the validity of the dataset, this paper proposes a teaching intonation assessment model (TIAM) based on low-level and deep-level features of speech. The experimental results show that TIAM based on the dataset constructed in this paper is basically consistent with the results of manual evaluation, and the results are better than the baseline models, which proves the effectiveness of the evaluation model.
△ Less
Submitted 14 December, 2023;
originally announced December 2023.
-
OpenGF: An Ultra-Large-Scale Ground Filtering Dataset Built Upon Open ALS Point Clouds Around the World
Authors:
Nannan Qin,
Weikai Tan,
Lingfei Ma,
Dedong Zhang,
Jonathan Li
Abstract:
Ground filtering has remained a widely studied but incompletely resolved bottleneck for decades in the automatic generation of high-precision digital elevation model, due to the dramatic changes of topography and the complex structures of objects. The recent breakthrough of supervised deep learning algorithms in 3D scene understanding brings new solutions for better solving such problems. However,…
▽ More
Ground filtering has remained a widely studied but incompletely resolved bottleneck for decades in the automatic generation of high-precision digital elevation model, due to the dramatic changes of topography and the complex structures of objects. The recent breakthrough of supervised deep learning algorithms in 3D scene understanding brings new solutions for better solving such problems. However, there are few large-scale and scene-rich public datasets dedicated to ground extraction, which considerably limits the development of effective deep-learning-based ground filtering methods. To this end, we present OpenGF, first Ultra-Large-Scale Ground Filtering dataset covering over 47 $km^2$ of 9 different typical terrain scenes built upon open ALS point clouds of 4 different countries around the world. OpenGF contains more than half a billion finely labeled ground and non-ground points, thousands of times the number of labeled points than the de facto standard ISPRS filtertest dataset. We extensively evaluate the performance of state-of-the-art rule-based algorithms and 3D semantic segmentation networks on our dataset and provide a comprehensive analysis. The results have confirmed the capability of OpenGF to train deep learning models effectively. This dataset is released at https://github.com/Nathan-UW/OpenGF to promote more advancing research for ground filtering and large-scale 3D geographic environment understanding.
△ Less
Submitted 12 April, 2021; v1 submitted 23 January, 2021;
originally announced January 2021.
-
Toronto-3D: A Large-scale Mobile LiDAR Dataset for Semantic Segmentation of Urban Roadways
Authors:
Weikai Tan,
Nannan Qin,
Lingfei Ma,
Ying Li,
Jing Du,
Guorong Cai,
Ke Yang,
Jonathan Li
Abstract:
Semantic segmentation of large-scale outdoor point clouds is essential for urban scene understanding in various applications, especially autonomous driving and urban high-definition (HD) mapping. With rapid developments of mobile laser scanning (MLS) systems, massive point clouds are available for scene understanding, but publicly accessible large-scale labeled datasets, which are essential for de…
▽ More
Semantic segmentation of large-scale outdoor point clouds is essential for urban scene understanding in various applications, especially autonomous driving and urban high-definition (HD) mapping. With rapid developments of mobile laser scanning (MLS) systems, massive point clouds are available for scene understanding, but publicly accessible large-scale labeled datasets, which are essential for developing learning-based methods, are still limited. This paper introduces Toronto-3D, a large-scale urban outdoor point cloud dataset acquired by a MLS system in Toronto, Canada for semantic segmentation. This dataset covers approximately 1 km of point clouds and consists of about 78.3 million points with 8 labeled object classes. Baseline experiments for semantic segmentation were conducted and the results confirmed the capability of this dataset to train deep learning models effectively. Toronto-3D is released to encourage new research, and the labels will be improved and updated with feedback from the research community.
△ Less
Submitted 16 April, 2020; v1 submitted 18 March, 2020;
originally announced March 2020.