Neural events compress event camera streams into fewer informative tokens via discrete asynchronous autoencoders, achieving on-par or better performance on detection and classification with 2x lower event rate.
hub Canonical reference
YOLOX: Exceeding YOLO Series in 2021
Canonical reference. 70% of citing Pith papers cite this work as background.
60
Pith papers citing it
Background
70% of classified citations
abstract
In this report, we present some experienced improvements to YOLO series, forming a new high-performance detector -- YOLOX. We switch the YOLO detector to an anchor-free manner and conduct other advanced detection techniques, i.e., a decoupled head and the leading label assignment strategy SimOTA to achieve state-of-the-art results across a large scale range of models: For YOLO-Nano with only 0.91M parameters and 1.08G FLOPs, we get 25.3% AP on COCO, surpassing NanoDet by 1.8% AP; for YOLOv3, one of the most widely used detectors in industry, we boost it to 47.3% AP on COCO, outperforming the current best practice by 3.0% AP; for YOLOX-L with roughly the same amount of parameters as YOLOv4-CSP, YOLOv5-L, we achieve 50.0% AP on COCO at a speed of 68.9 FPS on Tesla V100, exceeding YOLOv5-L by 1.8% AP. Further, we won the 1st Place on Streaming Perception Challenge (Workshop on Autonomous Driving at CVPR 2021) using a single YOLOX-L model. We hope this report can provide useful experience for developers and researchers in practical scenes, and we also provide deploy versions with ONNX, TensorRT, NCNN, and Openvino supported. Source code is at https://github.com/Megvii-BaseDetection/YOLOX.
hub tools
citation-role summary
background 7
method 2
baseline 1
citation-polarity summary
representative citing papers
NERVE is a new 600GB multi-sensor dataset with DVS, RGB-D, and 24/77GHz radar plus baselines showing DVS+77GHz radar fusion improves human detection to 47.5% mAP with sub-1.8m distance error.
CUTAL scores multi-frame clips for uncertainty and enforces temporal diversity to train transformer MOT models to near full-supervision performance with 50% of the labels.
LAMP tracks 3D human motion from moving multi-camera headsets by converting 2D detections to a unified metric 3D world frame via device localization and fitting with an end-to-end spatio-temporal transformer.
AniMatrix generates anime videos by structuring artistic production rules into a controllable taxonomy and training the model to prioritize those rules over physical realism, achieving top scores from professional animators on prompt understanding and artistic motion.
WUTDet is a 100K-image ship detection dataset with benchmarks indicating Transformer models outperform CNN and Mamba architectures in accuracy and small-object detection for complex maritime environments.
A million-box 159-class remote-sensing dataset plus a GSD- and hierarchy-aware detector yields ~5 mAP average gains over fully supervised baselines on nine external benchmarks with no target training.
Proposes DERNet with Decompose-Enhance-Reconstruct operator and three plug-and-play modules to shift small object detection from spatial to spectral feature processing, claiming better performance than YOLOv11 with 1/6 the parameters.
Polycepta recursively estimates per-object appearance states for tracking-by-detection MOT, improving identity preservation over time and reporting 92.27% MOTA on KITTI at 90.57 Hz.
FATE combines pillar encoding via orthogonal polynomial basis with frequency-aware training to enable event-based object detection at up to 200 Hz without internal temporal sub-binning.
Presents MMIO benchmark and RTVP method achieving state-of-the-art 42.2% AP in zero-shot industrial defect detection.
Presents MMIOC-1M benchmark with 1M+ samples across 14 super-categories and RTVPNet with domain projection, sparse sampling, and bidirectional interaction, claiming SOTA on MMIOC-1M, LVIS, and COCO.
Marker-free multi-camera AR guidance with synthetic-data 6D pose estimation reduces scrub nurse instrument assembly time by 21.3% and perceived workload versus paper manuals in knee arthroplasty simulations.
Presents a new egocentric HRI dataset and reports that an optimized tracking pipeline with ReID reduces identity switches by 49% over baseline.
E³C is a video diffusion model that disentangles persistent 3D scene structure via point-cloud memory from human dynamics via ego-exo pose controls for improved egocentric video generation on the Nymeria dataset.
A framework with new metrics and train-time/post-hoc calibrators aligns probabilistic object detectors to annotator disagreement distributions for classification and localization without ground truth.
TrajVAD shows that bounding-box trajectories modeled via normalizing flows can serve as a primary cue for video anomaly detection, with the trajectory-only variant achieving 87.7% AP on ShanghaiTech and best results on MSAD.
Contrastive pretraining on mammography atlas image-text pairs improves BI-RADS classification F1 by 1-14% especially in low-label regimes, outperforming equivalent numbers of direct labels in some settings.
SparseSAM achieves 2x faster inference and 2.8x memory reduction in SAM with only 0.004 mIoU loss at 0.4 density via Stripe-Sort Attention and Residual-Consistency MLP.
A deterministic queue-based matching algorithm using geometric overlaps and virtual lane discretization enables 99.8% handover success rate for continuous identity persistence in multi-UAV vehicle tracking.
CalibFree enables calibration-free multi-camera tracking via self-supervised feature separation through single-view distillation and cross-view reconstruction, reporting 3% higher accuracy and 7.5% better F1 on tested datasets.
FUN is an end-to-end Focal U-Net that performs joint hyperspectral image reconstruction and object detection via multi-task learning with focal modulation, achieving SOTA results with 40% fewer parameters and a new 363-image dataset.
GateMOT proposes Q-Gated Attention to enable linear-complexity, spatially aware attention for state-of-the-art dense object tracking on benchmarks like BEE24.
CAM3DNet outperforms prior camera-based 3D detectors on nuScenes, Waymo and Argoverse by using three new modules to better mine multi-scale spatiotemporal features from 2D queries and pyramid maps.
citing papers explorer
-
Neural Events: Discrete Asynchronous Autoencoders for Event-Based Vision
Neural events compress event camera streams into fewer informative tokens via discrete asynchronous autoencoders, achieving on-par or better performance on detection and classification with 2x lower event rate.
-
NERVE: A Neuromorphic Vision and Radar Ensemble for Multi-Sensor Fusion Research
NERVE is a new 600GB multi-sensor dataset with DVS, RGB-D, and 24/77GHz radar plus baselines showing DVS+77GHz radar fusion improves human detection to 47.5% mAP with sub-1.8m distance error.
-
Clip-level Uncertainty and Temporal-aware Active Learning for End-to-End Multi-Object Tracking
CUTAL scores multi-frame clips for uncertainty and enforces temporal diversity to train transformer MOT models to near full-supervision performance with 50% of the labels.
-
LAMP: Localization Aware Multi-camera People Tracking in Metric 3D World
LAMP tracks 3D human motion from moving multi-camera headsets by converting 2D detections to a unified metric 3D world frame via device localization and fitting with an end-to-end spatio-temporal transformer.
-
AniMatrix: An Anime Video Generation Model that Thinks in Art, Not Physics
AniMatrix generates anime videos by structuring artistic production rules into a controllable taxonomy and training the model to prioritize those rules over physical realism, achieving top scores from professional animators on prompt understanding and artistic motion.
-
WUTDet: A 100K-Scale Ship Detection Dataset and Benchmarks with Dense Small Objects
WUTDet is a 100K-image ship detection dataset with benchmarks indicating Transformer models outperform CNN and Mamba architectures in accuracy and small-object detection for complex maritime environments.
-
LEVIRDet: A Million-Scale 159-Category Dataset and Foundation Model for Universal Remote Sensing Object Detection
A million-box 159-class remote-sensing dataset plus a GSD- and hierarchy-aware detector yields ~5 mAP average gains over fully supervised baselines on nine external benchmarks with no target training.
-
From Spatial to Spectral: An Efficient, Frequency-Guided Feature Representation Learner for Small Object Detection
Proposes DERNet with Decompose-Enhance-Reconstruct operator and three plug-and-play modules to shift small object detection from spatial to spectral feature processing, claiming better performance than YOLOv11 with 1/6 the parameters.
-
Polycepta: Object-Centric Appearance Estimation for Multi-Object Tracking
Polycepta recursively estimates per-object appearance states for tracking-by-detection MOT, improving identity preservation over time and reporting 92.27% MOTA on KITTI at 90.57 Hz.
-
FATE: Pillar Encoding and Frequency-Aware Training for Event-Based Object Detection
FATE combines pillar encoding via orthogonal polynomial basis with frequency-aware training to enable event-based object detection at up to 200 Hz without internal temporal sub-binning.
-
Zero-Shot Learning in Industrial Scenarios: New Large-Scale Benchmark, Challenges and Baseline
Presents MMIO benchmark and RTVP method achieving state-of-the-art 42.2% AP in zero-shot industrial defect detection.
-
Unification of Closed-Open Industrial Detection Scenarios: New Large-Scale Benchmarks,Challenges and Baselines
Presents MMIOC-1M benchmark with 1M+ samples across 14 super-categories and RTVPNet with domain projection, sparse sampling, and bidirectional interaction, claiming SOTA on MMIOC-1M, LVIS, and COCO.
-
Multi-Camera AR Guidance System for Surgical Instrument Handling and Assembly: Investigating Workload and Efficiency
Marker-free multi-camera AR guidance with synthetic-data 6D pose estimation reduces scrub nurse instrument assembly time by 21.3% and perceived workload versus paper manuals in knee arthroplasty simulations.
-
Face versus Body Tracking for Human-Robot Interaction: An Egocentric Dataset
Presents a new egocentric HRI dataset and reports that an optimized tracking pipeline with ReID reduces identity switches by 49% over baseline.
-
E$^3$C: Video Generation with 3D Environmental Memory and Ego-Exo Human Pose Control
E³C is a video diffusion model that disentangles persistent 3D scene structure via point-cloud memory from human dynamics via ego-exo pose controls for improved egocentric video generation on the Nymeria dataset.
-
Calibrating Probabilistic Object Detectors with Annotator Disagreement
A framework with new metrics and train-time/post-hoc calibrators aligns probabilistic object detectors to annotator disagreement distributions for classification and localization without ground truth.
-
Bounding-Box Trajectories Matter for Video Anomaly Detection
TrajVAD shows that bounding-box trajectories modeled via normalizing flows can serve as a primary cue for video anomaly detection, with the trajectory-only variant achieving 87.7% AP on ShanghaiTech and best results on MSAD.
-
MAM-CLIP: Vision-Language Pretraining on Mammography Atlases for BI-RADS Classification
Contrastive pretraining on mammography atlas image-text pairs improves BI-RADS classification F1 by 1-14% especially in low-label regimes, outperforming equivalent numbers of direct labels in some settings.
-
SparseSAM: Structured Sparsification of Activations in Segment Anything Models
SparseSAM achieves 2x faster inference and 2.8x memory reduction in SAM with only 0.004 mIoU loss at 0.4 density via Stripe-Sort Attention and Residual-Consistency MLP.
-
A Topology-Aware Spatiotemporal Handover Framework for Continuous Multi-UAV Tracking
A deterministic queue-based matching algorithm using geometric overlaps and virtual lane discretization enables 99.8% handover success rate for continuous identity persistence in multi-UAV vehicle tracking.
-
CalibFree: Self-Supervised View Feature Separation for Calibration-Free Multi-Camera Multi-Object Tracking
CalibFree enables calibration-free multi-camera tracking via self-supervised feature separation through single-view distillation and cross-view reconstruction, reporting 3% higher accuracy and 7.5% better F1 on tested datasets.
-
FUN: A Focal U-Net Combining Reconstruction and Object Detection for Snapshot Spectral Imaging
FUN is an end-to-end Focal U-Net that performs joint hyperspectral image reconstruction and object detection via multi-task learning with focal modulation, achieving SOTA results with 40% fewer parameters and a new 363-image dataset.
-
GateMOT: Q-Gated Attention for Dense Object Tracking
GateMOT proposes Q-Gated Attention to enable linear-complexity, spatially aware attention for state-of-the-art dense object tracking on benchmarks like BEE24.
-
CAM3DNet: Comprehensively mining the multi-scale features for 3D Object Detection with Multi-View Cameras
CAM3DNet outperforms prior camera-based 3D detectors on nuScenes, Waymo and Argoverse by using three new modules to better mine multi-scale spatiotemporal features from 2D queries and pyramid maps.
-
Improving Layout Representation Learning Across Inconsistently Annotated Datasets via Agentic Harmonization
VLM-based harmonization of inconsistent annotations across two document layout corpora raises detection F-score from 0.860 to 0.883 and table TEDS from 0.750 to 0.814 while tightening embedding clusters.
-
Scale-Gest: Scalable Model-Space Synthesis and Runtime Selection for On-Device Gesture Detection
Scale-Gest creates a runtime-selectable family of tiny-YOLO models with device-calibrated ACE profiles and an ROI gate that cuts per-frame energy by 4x while holding event-level F1 at 0.8-0.9 on a new driving-gesture dataset.
-
Focus on What Really Matters in Low-Altitude Governance: A Management-Centric Multi-Modal Benchmark with Implicitly Coordinated Vision-Language Reasoning Framework
Presents the first management-oriented multi-modal benchmark GovLA-10K and a vision-language reasoning framework GovLA-Reasoner with a spatially-aware adapter for low-altitude aerial perception.
-
AHCQ-SAM: Toward Accurate and Hardware-Compatible Post-Training Segment Anything Model Quantization
AHCQ-SAM introduces ACNR, HLUQ, CAG, and LNQ quantization techniques that deliver 15.2% mAP gain on 4-bit SAM-B and 14.01% J&F gain on 4-bit SAM2-Tiny versus prior PTQ methods.
-
Dual-Head Knowledge Distillation: Enhancing Logits Utilization with an Auxiliary Head
Dual-head knowledge distillation partitions the linear classifier into separate heads for logit and probability losses to exploit logits without causing classification head collapse.
-
DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection
DINO reaches 51.3 AP on COCO val2017 with a ResNet-50 backbone after 24 epochs, a +2.7 AP gain over the prior best DETR variant.
-
PS-MOT: Cultivating Instance Awareness from Point Seeds for Multi-Object Tracking
PS-Track sets a new state-of-the-art for point-supervised multi-object tracking by converting point seeds into temporally consistent pseudo-labels via Temporal-Feedback Prompting, Point-Excited Wavelet Attention, and Uncertainty-Guided Gaussian Learning.
-
Efficient RGB-T Object Detection via Sparse Cross-Modality Fusion
A two-stage RGB-T detector performs lightweight modality-specific proposal generation followed by sparse fusion-based refinement to match accuracy of heavier models at lower parameter and compute cost.
-
Kairos: A Regret-Aware Native World-Action Model Stack for Physical AI
Kairos learns and maintains control-sufficient world states via a cross-embodiment curriculum, hybrid linear temporal attention, and deployment-aware co-design for Physical AI.
-
Isolation-aware Scheduling Framework for DNN-based End-to-End Autonomous Driving System on Tile-based Accelerators
ADS-Tile is a new scheduling framework using spatio-temporal isolation-sharing and a DAG-aware probabilistic model to cut tile usage by up to 32% and reallocation waste to under 1.2% for E2E ADS DNN pipelines on tile accelerators.
-
FDDet: Achieving Data-Efficient Food Defect Detection Under Real-World Scenarios
FDDet is a semi-supervised object detection framework with BBoxMixUp and CGPC that outperforms standard detectors on the new FDD-48 food defect dataset under data-limited real-world conditions.
-
Dual-Integrated Low-Latency Single-Lens Infrared Computational Imaging for Object Detection
PDI-Net integrates a semi-U-Net encoder with YOLO detection using a physics-aware PALS-Bridge and optical simulation to deliver 84% faster inference and 5% higher mAP than pruned reconstruction-plus-detection on low-SNR M3FD infrared data.
-
STAR-IOD: Scale-decoupled Topology Alignment with Pseudo-label Refinement for Remote Sensing Incremental Object Detection
STAR-IOD applies scale-decoupled topology alignment and K-Means-based pseudo-label refinement to reduce catastrophic forgetting in remote sensing incremental object detection, reporting 1.7% and 2.1% mAP gains on new DIOR-IOD and DOTA-IOD datasets.
-
MR2-ByteTrack: CNN and Transformer-based Video Object Detection for AI-augmented Embedded Vision Sensor Nodes
MR2-ByteTrack maintains high accuracy in video object detection on MCUs by combining multi-resolution processing, ByteTrack for frame linking, and Rescore for confidence aggregation, achieving up to 55% energy savings and real-time performance for both CNN and Transformer models.
-
Portable Active Learning for Object Detection
PAL is a portable active learning method for object detection that uses class-specific logistic classifiers for uncertainty and image-level diversity to select annotation batches, showing better label efficiency than baselines on COCO, VOC, and BDD100K.
-
Utility-Aware Progressive Inference over UDP Packet Blocks for Emergency Communications
Utility-aware progressive inference on UDP packet blocks enables early hazard recognition, reducing packet budget by 34.2% and decision delay by 1209 ms while retaining 91.5% of full-reception accuracy.
-
SAMOFT: Robust Multi-Object Tracking via Region and Flow
SAMOFT improves multi-object tracking by using SAM segmentation and optical flow for pixel-level motion matching, flexible centroid correction, and training-free motion pattern fixes on top of standard Kalman and ReID baselines.
-
Time-series Meets Complex Motion Modeling: Robust and Computational-effective Motion Predictor for Multi-object Tracking
TCMP achieves SOTA MOT metrics (HOTA 63.4%, IDF1 65.0%, AssA 49.1%) with 0.014x parameters and 0.05x FLOPs of the previous best method by using a simple dilated TCN regressor.
-
SocialMirror: Reconstructing 3D Human Interaction Behaviors from Monocular Videos with Semantic and Geometric Guidance
SocialMirror reconstructs 3D meshes of closely interacting humans from monocular videos using semantic guidance from vision-language models and geometric constraints in a diffusion model to handle occlusions and maintain temporal and spatial consistency.
-
Hypergraph-State Collaborative Reasoning for Multi-Object Tracking
HyperSSM integrates hypergraphs and state space models to let correlated objects mutually refine motion estimates, stabilizing trajectories under noise and occlusion for state-of-the-art multi-object tracking.
-
Empirical Evaluation of PDF Parsing and Chunking for Financial Question Answering with RAG
Systematic tests show that specific PDF parsers combined with overlapping chunking strategies better preserve structure and improve RAG answer correctness on financial QA benchmarks including the new TableQuest dataset.
-
Enhancing Event-based Object Detection with Monocular Normal Maps
NRE-Net adds geometric priors from RGB-derived normal maps to RGB and event data via ADFM and EAFM fusion modules, reporting 3% AP50 gains over dual-modal baselines on driving datasets.
-
Automatic Road Subsurface Distress Recognition from Ground Penetrating Radar Images using Deep Learning-based Cross-verification
A cross-verification strategy using three YOLO models trained on distinct views of a 2134-sample 3D GPR dataset detects road subsurface distress with over 98.6 percent recall on field data.
-
HunyuanVideo: A Systematic Framework For Large Video Generative Models
HunyuanVideo presents a 13B-parameter open-source video generative model with integrated data, architecture, training, and inference systems whose professional evaluations show it outperforming prior SOTA models including Runway Gen-3 and Luma 1.6.
-
MLT-Dedup: Efficient Large-Scale Online Video Deduplication via Multi-Level Representations and Spatial-Temporal Matching
MLT-Dedup achieves 91% reduction in online video repetition rates at 90% precision and 5x indexing capacity using multi-level representations and differential feature-enhanced similarity on a real-world platform.
-
Optimizing Latent Representations for Robust Building Damage Assessment Onboard Earth Observation Satellites
Develops and benchmarks a siamese latent-representation system for object-level building damage assessment directly onboard Earth observation satellites.