YOLOv14 is presented, a unified adaptive detection framework that addresses real-time object detectors variations through four complementary mechanisms, formalized under a novel Adaptive Routing and Modulation (ARM) paradigm.
Abstract
Real-time object detectors achieve remarkable accuracy under controlled conditions, yet degrade sharply on non-ideal inputs-fisheye distortion, game-rendered content, aerial views, and 360{\deg}panoramas. We present YOLOv14, a unified adaptive detection framework that addresses these variations through four complementary mechanisms, formalized under a novel Adaptive Routing and Modulation (ARM) paradigm. Unlike conventional unsupervised domain adaptation, our approach employs Target-Prior Guided Source-Domain Augmentation(TP-SDA), using only 50 unlabeled target images offline to estimate style statistics, while adversarial alignment serves as a lightweight regularizer rather than the primary adaptation driver. Together, these components enable YOLOv14 to achieve 49.1 mAP on COCO val2017 at 2.91 ms (T4 GPU), with substantial gains of +4.1 (fisheye), +6.6 (panorama), +6.4 (drone), and +26.1 (gamestylized) mAP over YOLOv12s. Crucially, we validate generalization on real-world game screenshots (GTA-V, Unity), achieving +14.2 mAP, confirming practical transferability beyond synthetic benchmarks. Code and models are released at https://github.com/zhangcbb/yolov14.
Unsupervised Domain Adaptation (UDA) for object detection remains challenging under adverse weather due to significant distribution shifts. While recent Vision Foundation Model (VFM) based methods show promise, they often encounter limitations in extreme domain gaps and pseudo-label noise. This paper proposes two enhancements to the DINO Teacher framework: (1) a multi-round self-training strategy to refine the labeling model progressively, and (2) a depth-guided spatial modulation mechanism using geometric priors from a DINOv3based depth estimator. By modulating the input space, the student model is encouraged to emphasize spatial cues that are less sensitive to visibility degradation in foggy environments. Experiments on Foggy Cityscapes demonstrate that our approach reaches 56.5% mAP with a VGG-16 backbone and 59.8% mAP with ResNet-50. These results demonstrate competitive performance compared with the DINO Teacher baseline and recent Vision Language Model (VLM) based methods, particularly for tail categories such as bus and train.
T. Doan, D. C. Bui, Khanh-Duy Nguyen et al.· International Conference on...· 0 citations
Small object detection in crowded urban and aerial scenes remains a critical challenge due to limited pixel information and information loss in deep neural networks. This study introduces a novel optimization framework for YOLOv11, specifically engineered for tiny-scale targets by integrating convolutional block attention modules (CBAM), k-means anchor clustering, and an enhanced feature pyramid network (FPN). Evaluated on the TinyPerson and COCO-mini datasets, the YOLOv11-optimized model achieves significant performance breakthroughs, delivering a +7.3% gain in mean average precision (mAP) and a +10.5% increase in recall over the baseline. Notably, the model achieved a recall of 0.072 on the TinyPerson dataset, with double sensitivity of standard YOLOv11. With a high-speed inference rate of 27.3 FPS, this research demonstrates that strategic architectural refinements can drastically improve small object detection reliability without compromising real-time viability on edge devices.
Husna Sarirah Husin, H. Hao, Yuan-Fei Pan et al.· IAES International Journal o...· 0 citations
Real-time object detection needs to strike a balance between accuracy, speed, and reliability. However,many YOLO-based models struggle with two main issues: weakened feature quality in deeper backbonelayers and inefficient multiscale fusion in the detection neck. To overcome these challenges, we presentEnYOLO-World — a refined detection framework that combines the YOLOv9 GELAN backbone withProgrammable Gradient Information (PGI) to retain detailed spatial features and ensure smoother gradi-ent flow. We also introduce a FiLM-Driven Path Aggregation Network (FiLM-PAN), which adaptivelyfuses multiscale features using channel-wise affine transformations, achieving efficient performance withlinear complexity. Tests on the COCO dataset show that EnYOLO-World offers a better trade-off betweenaccuracy and speed compared to existing real-time detectors, demonstrating the value of optimizing bothbackbone feature retention and neck-level fusion strategies.
Streaming 3D reconstruction demands both speed and temporal fidelity, goals that existing methods undermine by updating every Gaussian every frame, even in static regions. We present DecoGS, a method for efficient online training of 3D Gaussians from streaming videos. Unlike prior methods that update the entire scene indiscriminately, DecoGS introduces an adaptive mechanism that selectively focuses optimization on spatiotemporal regions exhibiting motion or photometric changes. This targeted training strategy eliminates redundant updates that cause flickering and drift in nominally static regions, while enabling fast, high-fidelity scene updates. The pipeline further integrates region-aware Gaussian management through gradient gating and efficient visibility filtering to maintain temporal coherence and a compact memory footprint. On N3DV and MeetRoom, DecoGS achieves 34.55 and 31.60 dB PSNR respectively, outperforming all streaming and offline baselines, while rendering at 261 FPS with $70\times$ lower temporal flicker than the best prior method, requiring no large-scale pretraining.
Idil Sulo, Alexey Supikov, Ilke Demir et al.· 0 citations
PoseAdapter, a lightweight framework for high-fidelity 2.5D controllable image generation, and a Context-Aware Dual-Stream Representation, to resolve the generative trade-off between strict instance isolation and global coherence.
Yu-Feng Chi, Hui-Min Ma, Fan Gao et al.· 0 citations
Object detectors often degrade under domain shifts such as changes in lighting, weather, or occlusion. These shifts alter object appearance and expose a reliance on visual shortcuts learned from the training distribution that do not generalize across domains. Acquiring sufficient real-world samples to capture such domain variation is particularly difficult in specialized, low-data settings. Recent advances in diffusion-based generative image editing have shown promise for improving the in-domain performance of object detectors through synthetic data augmentation. However, their potential to improve out-of-domain robustness remains largely unexplored. We hypothesize that generative image editing can simulate a controlled domain shift in training data, effectively bridging the gap between source and target domains. To test this, we studied camouflaged military vehicle detection as a challenging domain shift scenario. Detectors trained on uncamouflaged data demonstrate substantial degradation on real test imagery containing foliage, netting, and multi-spectral camouflage across 15 vehicle classes in close-up, ground-level imagery. We used two diffusion-based editing models, Qwen Image Edit 2509 and Flux.2 Dev, to synthetically add camouflage to the training data, alongside a LoRA fine-tuned version of Qwen. A non-generative black-bar occlusion baseline served as a lower bound on augmentation quality. Using a GroundingDINO detector trained on real and synthetic data, generative camouflage augmentation yielded substantial mAP improvements for foliage (+20.1) and netting (+14.4) camouflage. Generating multi-spectral camouflage proved more challenging, but LoRA fine-tuning improved performance by 4.4 mAP over the uncamouflaged baseline.
Isabel D. Stein, Thijs A. Eker, Sebastiaan P. Snel et al.· Artificial Intelligence for...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
