Introducing clDice, a topology-preserving loss for tubular structure segmentation, improving connectivity and graph similarity in 2D/3D data.

Suprosanna Shit, Johannes C. Paetzold, Anjany Sekuboyina et al.

Proposed a joint deconvolution and deep learning framework for UDC image restoration, achieving PSNR 28.5dB and 20fps real-time performance.

Yuqian Zhou, David Ren, Neil Emerton et al.

Convolutional Occupancy Networks (CON) combine convolutional encoders with implicit occupancy decoders, achieving high-precision 3D reconstruction of complex scenes.

Songyou Peng, Michael Niemeyer, Lars Mescheder et al.

StyleGAN2 distillation method enables feed-forward image manipulation with quality comparable to StyleGAN2 backpropagation.

Yuri Viazovetskyi, Vladimir Ivashkin, Evgeny Kashin

Proposed IF-Nets use multi-scale deep feature tensors for continuous, multi-topology 3D shape reconstruction, outperforming prior methods in detail preservation and articulated human modeling.

Julian Chibane, Thiemo Alldieck, Gerard Pons-Moll

Learn2Perturb enhances adversarial robustness through end-to-end feature perturbation learning, achieving 4-7% improvement on CIFAR-10.

Ahmadreza Jeddi, Mohammad Javad Shafiee, Michelle Karg et al.

SAT combines state-aware feedback for real-time VOS, reaching 72.3% J&F and 39 FPS on DAVIS2017-Val.

Xi Chen, Zuoxin Li, Ye Yuan et al.

Cops-Ref dataset introduces complex logical reasoning for referring expression comprehension; current models perform poorly, highlighting room for improvement.

Zhenfang Chen, Peng Wang, Lin Ma et al.

Proposes a fast, robust VOS method combining a Gauss-Newton optimized target model with a U-Net segmentation network, achieving 41 FPS and state-of-the-art accuracy.

Andreas Robinson, Felix Järemo Lawin, Martin Danelljan et al.

Defense-PointNet employs a discriminator to enhance PointNet's robustness against FGSM adversarial point clouds, outperforming baseline defenses with 96.35% accuracy.

Yu Zhang, Gongbo Liang, Tawfiq Salem et al.

Proposed generalized ODIN achieves high OoD detection without out-of-distribution data tuning, via confidence decomposition and input preprocessing.

Yen-Chang Hsu, Yilin Shen, Hongxia Jin et al.

Proposes a pre-training + fine-tuning paradigm using image-text-action triplets, improving success rate from 47% to 51% in VLN tasks.

Weituo Hao, Chunyuan Li, Xiujun Li et al.

Proposes a background modeling approach within a distillation framework, achieving 69.6% mIoU on Pascal-VOC 2012 in incremental semantic segmentation.

Fabio Cermelli, Massimiliano Mancini, Samuel Rota Bulò et al.

Proposes Gram-Net, leveraging global texture features to enhance robustness and generalization in deepfake face detection.

Zhengzhe Liu, Xiaojuan Qi, Philip Torr

This paper introduces a self-supervised framework embedding a differentiable point-based motion estimator to learn robust radar keypoints, reducing odometry errors by 45% and increasing speed tenfold.

Dan Barnes, Ingmar Posner

Enhanced Virtual KITTI 2 with Unity HDRP, stereo camera, and weather variations; validated with deep learning models showing high transferability.

Yohann Cabon, Naila Murray, Martin Humenberger

Proposes MM-SADA, combining multi-modal self-supervision and adversarial training, improving cross-domain fine-grained action recognition by 4.8%.

Jonathan Munro, Dima Damen

PLR-OSNet combines global and part-level features with a single ID loss, achieving 88.9% mAP and 95.6% Rank-1 on Market1501 with only 3.4M parameters.

Ben Xie, Xiaofu Wu, Suofei Zhang et al.

RSL-Net uses radar and satellite imagery to achieve vehicle localization without prior maps, with mean angular error ~3°, translation ~3m.

Tim Y. Tang, Daniele De Martini, Dan Barnes et al.

DeeperForensics-1.0 is a large-scale face forgery dataset with 60,000 videos, 10x larger than existing datasets, built using a novel DF-VAE framework to enhance realism and diversity.

Liming Jiang, Ren Li, Wayne Wu et al.