OD3 presents an optimization-free dataset distillation framework for object detection that reports new state-of-the-art accuracy on COCO and VOC at compression ratios from 0.25% to 5%.
hub Mixed citations
Focal Loss for Dense Object Detection
Mixed citation behavior. Most common role is method (50%).
39
Pith papers citing it
1
external citations · Pith
Method
50% of classified citations
abstract
The highest accuracy object detectors to date are based on a two-stage approach popularized by R-CNN, where a classifier is applied to a sparse set of candidate object locations. In contrast, one-stage detectors that are applied over a regular, dense sampling of possible object locations have the potential to be faster and simpler, but have trailed the accuracy of two-stage detectors thus far. In this paper, we investigate why this is the case. We discover that the extreme foreground-background class imbalance encountered during training of dense detectors is the central cause. We propose to address this class imbalance by reshaping the standard cross entropy loss such that it down-weights the loss assigned to well-classified examples. Our novel Focal Loss focuses training on a sparse set of hard examples and prevents the vast number of easy negatives from overwhelming the detector during training. To evaluate the effectiveness of our loss, we design and train a simple dense detector we call RetinaNet. Our results show that when trained with the focal loss, RetinaNet is able to match the speed of previous one-stage detectors while surpassing the accuracy of all existing state-of-the-art two-stage detectors. Code is at: https://github.com/facebookresearch/Detectron.
hub tools
citation-role summary
background 4
method 4
citation-polarity summary
representative citing papers
VitaminP uses paired H&E-mIF data to train a model that transfers molecular boundary information, enabling accurate whole-cell segmentation directly from routine H&E histology across 34 cancer types.
TCG-AR is a real-time multi-view AR system for trading card games using only commodity RGB cameras and synthetic training data.
Grad Detect uses internal gradient patterns from one inference pass to predict LLM hallucinations and abstention, outperforming confidence and sampling baselines on Q&A benchmarks with most signal in the final five layers.
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.
Develops differentiable temporally weighted score-oriented losses that incorporate event timing to better match training with time-localized detection utilities on benchmark datasets.
A geometry-aware dynamic-query transformer decoder with Local Strided Cross-Attention raises track reconstruction efficiency from 94.1% to 98.1%, halves latency, and cuts memory use by over 10x versus fixed-query baselines in a simplified HL-LHC simulation.
TaRO improves video temporal grounding in MLLMs via constructive reasoning exploration from dense captions and a temporal-sensitivity reward that uses logit drops on disrupted event boundaries, followed by curriculum learning to SOTA results.
AnyScene is an occupancy-centric framework using a Spatial-Temporal Occupancy Diffusion Transformer and Geometry-Grounded View Expansion to generate controllable driving scenes and videos from BEV layouts.
SGD is reformulated via a master equation from discrete updates, producing a discrete Fokker-Planck equation that predicts non-stationary variance growth proportional to learning rate in flat Hessian directions.
SToRe3D delivers up to 3x faster inference for multi-view 3D object detection in ViTs by selecting relevant 2D tokens and 3D queries via mutual relevance heads with only marginal accuracy loss.
Optimus mitigates toxicity during LLM fine-tuning by combining repurposed LLM safety alignments for detection with synthetic data and DPO alignment, remaining effective even with highly biased classifiers and against attacks.
GigaCheck detects LLM-generated text at both document and span levels by combining fine-tuned language-model embeddings with a DETR-like architecture that treats generated intervals as detectable objects.
A spectral vision transformer achieves equitable or superior performance with fewer parameters than standard ViTs, CNNs, and other models by using spectral projections for tokenization in limited-data medical imaging.
UniISP unifies ISP processing with a Hybrid Attention Module and Feature Adapter to produce images that are both visually pleasing for humans and informative for computer vision models.
TOFU loss mitigates the narrowing of generative diversity in LLMs after supervised fine-tuning by addressing neglect of low-frequency patterns and forgetting of prior knowledge.
CATMIL augments nnU-Net with component-adaptive Tversky and MIL-based lesion supervision to raise Dice scores, small-lesion recall, and error control on the MSLesSeg dataset.
LAA-X uses multi-task learning with explicit localized artifact attention and blending synthesis to build a deepfake detector that generalizes to high-quality and unseen manipulations after training only on real and pseudo-fake samples.
SPAR is a street-legal physical rim that cuts modern ALPR accuracy by 60% and reaches 18% targeted impersonation while costing under $100 and requiring no plate modification.
NORA applies task-aware weighting and NPK filtering to handle label noise in multi-attribute tagging of financial numerical entities, outperforming baselines on a new 6.6M-instance benchmark.
Replacement Learning replaces selected blocks in CNNs and ViTs with learnable parameter-fusion surrogates derived from adjacent layers to reduce full-depth backpropagation redundancy.
CUD reshapes the teacher's predictive distribution before distillation so that students receive calibrated uncertainty signals alongside accuracy, yielding more robust and better-calibrated models on high-cardinality and distribution-shift benchmarks.
LiLAW learns to weight samples as easy, moderate or hard using three global scalars updated by one gradient step on a validation batch to improve noisy training performance.
DeepFWI is a multi-modal LSTM model with cross-attention that identifies bug-sensitive warnings at warning granularity, reaching 67.06% F1 on a 280k-warning dataset and surfacing 25 confirmed bugs in four open-source projects.
citing papers explorer
-
OD3: Optimization-free Dataset Distillation for Object Detection
OD3 presents an optimization-free dataset distillation framework for object detection that reports new state-of-the-art accuracy on COCO and VOC at compression ratios from 0.25% to 5%.
-
VitaminP: cross-modal learning enables whole-cell segmentation from routine histology
VitaminP uses paired H&E-mIF data to train a model that transfers molecular boundary information, enabling accurate whole-cell segmentation directly from routine H&E histology across 34 cancer types.
-
TCG-AR: Real-Time Multi-View Augmented Reality for Trading Card Game Streaming
TCG-AR is a real-time multi-view AR system for trading card games using only commodity RGB cameras and synthetic training data.
-
Grad Detect: Gradient-Based Hallucination Detection in LLMs
Grad Detect uses internal gradient patterns from one inference pass to predict LLM hallucinations and abstention, outperforming confidence and sampling baselines on Q&A benchmarks with most signal in the final five layers.
-
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.
-
Weighted Score-Oriented Losses for Temporally Localized Event Prediction
Develops differentiable temporally weighted score-oriented losses that incorporate event timing to better match training with time-localized detection utilities on benchmark datasets.
-
Better Queries, Cheaper Attention: Adapting Transformers for Efficient Sparse Reconstruction
A geometry-aware dynamic-query transformer decoder with Local Strided Cross-Attention raises track reconstruction efficiency from 94.1% to 98.1%, halves latency, and cuts memory use by over 10x versus fixed-query baselines in a simplified HL-LHC simulation.
-
Temporal-Aware Reasoning Optimization for Video Temporal Grounding
TaRO improves video temporal grounding in MLLMs via constructive reasoning exploration from dense captions and a temporal-sensitivity reward that uses logit drops on disrupted event boundaries, followed by curriculum learning to SOTA results.
-
AnyScene: Towards Highly Controllable Driving Scene Generation at Anywhere and Beyond
AnyScene is an occupancy-centric framework using a Spatial-Temporal Occupancy Diffusion Transformer and Geometry-Grounded View Expansion to generate controllable driving scenes and videos from BEV layouts.
-
Why SGD is not Brownian Motion: A New Perspective on Stochastic Dynamics
SGD is reformulated via a master equation from discrete updates, producing a discrete Fokker-Planck equation that predicts non-stationary variance growth proportional to learning rate in flat Hessian directions.
-
SToRe3D: Sparse Token Relevance in ViTs for Efficient Multi-View 3D Object Detection
SToRe3D delivers up to 3x faster inference for multi-view 3D object detection in ViTs by selecting relevant 2D tokens and 3D queries via mutual relevance heads with only marginal accuracy loss.
-
Optimus: A Robust Defense Framework for Mitigating Toxicity while Fine-Tuning Conversational AI
Optimus mitigates toxicity during LLM fine-tuning by combining repurposed LLM safety alignments for detection with synthetic data and DPO alignment, remaining effective even with highly biased classifiers and against attacks.
-
GigaCheck: Detecting LLM-generated Content via Object-Centric Span Localization
GigaCheck detects LLM-generated text at both document and span levels by combining fine-tuned language-model embeddings with a DETR-like architecture that treats generated intervals as detectable objects.
-
Spectral Vision Transformer for Efficient Tokenization with Limited Data
A spectral vision transformer achieves equitable or superior performance with fewer parameters than standard ViTs, CNNs, and other models by using spectral projections for tokenization in limited-data medical imaging.
-
UniISP: A Unified ISP Framework for Both Human and Machine Vision
UniISP unifies ISP processing with a Hybrid Attention Module and Feature Adapter to produce images that are both visually pleasing for humans and informative for computer vision models.
-
Diversity in Large Language Models under Supervised Fine-Tuning
TOFU loss mitigates the narrowing of generative diversity in LLMs after supervised fine-tuning by addressing neglect of low-frequency patterns and forgetting of prior knowledge.
-
Component-Adaptive and Lesion-Level Supervision for Improved Small Structure Segmentation in Brain MRI
CATMIL augments nnU-Net with component-adaptive Tversky and MIL-based lesion supervision to raise Dice scores, small-lesion recall, and error control on the MSLesSeg dataset.
-
LAA-X: Unified Localized Artifact Attention for Quality-Agnostic and Generalizable Face Forgery Detection
LAA-X uses multi-task learning with explicit localized artifact attention and blending synthesis to build a deepfake detector that generalizes to high-quality and unseen manipulations after training only on real and pseudo-fake samples.
-
Street-Legal Physical-World Adversarial Rim for License Plates
SPAR is a street-legal physical rim that cuts modern ALPR accuracy by 60% and reaches 18% targeted impersonation while costing under $100 and requiring no plate modification.
-
Noise-Robust Financial Numerical Entity Attribute Tagging
NORA applies task-aware weighting and NPK filtering to handle label noise in multi-attribute tagging of financial numerical entities, outperforming baselines on a new 6.6M-instance benchmark.
-
Replacement Learning: Training Neural Networks with Fewer Parameters
Replacement Learning replaces selected blocks in CNNs and ViTs with learnable parameter-fusion surrogates derived from adjacent layers to reduce full-depth backpropagation redundancy.
-
Trust the uncertain teacher: distilling dark knowledge via calibrated uncertainty
CUD reshapes the teacher's predictive distribution before distillation so that students receive calibrated uncertainty signals alongside accuracy, yielding more robust and better-calibrated models on high-cardinality and distribution-shift benchmarks.
-
LiLAW: Lightweight Learnable Adaptive Weighting to Learn Sample Difficulty & Improve Noisy Training
LiLAW learns to weight samples as easy, moderate or hard using three global scalars updated by one gradient step on a validation batch to improve noisy training performance.
-
DeepFWI: Identifying Bug-Sensitive Warnings with Multi-Modal Code-Warning Semantics
DeepFWI is a multi-modal LSTM model with cross-attention that identifies bug-sensitive warnings at warning granularity, reaching 67.06% F1 on a 280k-warning dataset and surfacing 25 confirmed bugs in four open-source projects.
-
MapATM: Enhancing HD Map Construction through Actor Trajectory Modeling
MapATM improves lane divider AP by 4.6 and mAP by 2.6 on NuScenes by treating actor trajectories as structural priors for road geometry.
-
Low-Data Supervised Adaptation Outperforms Prompting for Cloud Segmentation Under Domain Shift
Supervised fine-tuning with 0.1% labeled data outperforms all 60 tested prompt variants for CLIPSeg cloud segmentation on satellite imagery under domain shift.
-
A Weak-Signal-Aware Framework for Subsurface Defect Detection: Mechanisms for Enhancing Low-SCR Hyperbolic Signatures
WSA-Net uses partial convolutions, heterogeneous grouping attention, geometric reconstruction, and context anchoring to enhance low-SCR hyperbolic signatures in GPR data, reaching 0.6958 mAP@0.5 at 164 FPS with 2.412M parameters on the RTST dataset.
-
Sequential Feature Selection for Efficient Landslide Segmentation from Multi-Spectral Data
SFFS yields an 8-channel subset matching or exceeding up to 30-channel landslide segmentation F1 on Landslide4Sense via iterative build-and-prune with U-Net++ proxy.
-
OneSearch-V2: The Latent Reasoning Enhanced Self-distillation Generative Search Framework
OneSearch-V2 improves generative retrieval via latent reasoning and self-distillation, achieving +3.98% item CTR, +2.07% buyer volume, and +2.11% order volume in online A/B tests.
-
A Resource Efficient Fusion Network for Object Detection in Bird's-Eye View using Camera and Raw Radar Data
Describes a camera-radar fusion network that uses raw RD spectra and BEV-polar camera features for BEV object detection, evaluated for accuracy and compute on the RADIal dataset.
-
Skywork-Reward: Bag of Tricks for Reward Modeling in LLMs
Data-centric filtering yields an 80K preference dataset and reward models that lead RewardBench while boosting other top entries.
-
Lung Nodules Detection and Segmentation Using 3D Mask-RCNN
Adapted Mask-RCNN to 3D and applied it to lung nodule detection and segmentation on CT scans, reporting competitive detection results on the LUNA16 dataset.
-
REFNet++: Multi-Task Efficient Fusion of Camera and Radar Sensor Data in Bird's-Eye Polar View
REFNet++ aligns raw camera images and radar range-Doppler data into a shared bird's-eye polar view using variational encoders for multi-task vehicle detection and free space segmentation on the RADIal dataset.
-
YOLOv3: An Incremental Improvement
YOLOv3 achieves accuracy comparable to SSD and RetinaNet but runs substantially faster, with 28.2 mAP at 320x320 in 22 ms and 57.9 mAP@50 in 51 ms on Titan X.
-
AI-Driven Security Alert Screening and Alert Fatigue Mitigation in Security Operations Centers: A Comprehensive Survey
A literature survey synthesizes 119 studies on AI-driven alert screening into a four-stage taxonomy of filtering, triage, correlation, and generative augmentation while identifying gaps in deployment realism and robustness.
-
Duluth at SemEval-2026 Task 6: DeBERTa with LLM-Augmented Data for Unmasking Political Question Evasions
DeBERTa-V3-base with focal loss, discourse features, and LLM-augmented data for minority classes achieves 0.76 Macro F1 on clarity-level classification of political QA pairs, ranking 8th in SemEval-2026 Task 6.
-
SoccerNet 2026 Player-Centric Ball-Action Spotting:Retraining and Post-Processing Extensions to the FOOTPASS Baselines
Four extensions to FOOTPASS baselines for player-centric ball-action spotting in soccer videos achieve 0.548 Macro F1 on test set and 0.446 on challenge set.
-
YEZE at SemEval-2026 Task 9: Detecting Multilingual, Multicultural and Multievent Online Polarization via Heterogeneous Ensembling
A heterogeneous ensemble of XLM-RoBERTa-large and mDeBERTa-v3-base with independent task modeling and class weighting is reported as effective for multilingual, multicultural, and multievent online polarization detection.
- MalariAI: A Label-Resilient Decoupled Framework for Annotation-Agnostic Cell Segmentation and Explainable Stage Classification in Dense Malaria Blood Smears