ProcNets leverages a large-scale YouCook2 dataset to achieve unsupervised procedure segmentation in long videos, outperforming baselines with 52% Jaccard and 48% mIoU.

Luowei Zhou, Chenliang Xu, Jason J. Corso

Deep appearance features integrated into SORT reduce identity switches by 45%, enhancing long-term occlusion tracking.

Nicolai Wojke, Alex Bewley, Dietrich Paulus

Proposed a deep learning-based image matting algorithm, achieving state-of-the-art alpha matte accuracy in complex scenes.

Ning Xu, Brian Price, Scott Cohen et al.

Proposed Global Convolutional Network (GCN) improves semantic segmentation, achieving 82.2% on VOC2012 and 76.9% on Cityscapes.

Chao Peng, Xiangyu Zhang, Gang Yu et al.

Combining self-supervised learning and imitation, the robot successfully manipulates ropes using 60K interaction data.

Ashvin Nair, Dian Chen, Pulkit Agrawal et al.

GPU-based k-selection algorithm achieves 55% of peak performance, 8.5x faster than prior art.

Jeff Johnson, Matthijs Douze, Hervé Jégou

Probabilistic graphical models enable automatic scientific poster generation by learning content layout and graphical element arrangement.

Yu-ting Qiang, Yanwei Fu, Xiao Yu et al.

Proposes ADDA, combining discriminative models with unshared weights and GAN loss for unsupervised domain adaptation, outperforming SOTA.

Eric Tzeng, Judy Hoffman, Kate Saenko et al.

Introduces CMP, an end-to-end deep learning architecture combining spatial mapping and differentiable planning for visual navigation.

Saurabh Gupta, Varun Tolani, James Davidson et al.

Proposed a semi-supervised deep learning method combining sparse LiDAR data and stereo images, achieving SOTA performance on KITTI dataset for monocular depth prediction.

Yevhen Kuznietsov, Jörg Stückler, Bastian Leibe

Estimate relative camera pose using CNNs, outperforming SURF and ORB.

Iaroslav Melekhov, Juha Ylioinas, Juho Kannala et al.

Proposes an adaptive robust loss function based on probabilistic modeling, improving vision task performance.

Jonathan T. Barron

Proposes an automated CNN recommendation system using task complexity and model ability scores, enabling fast, training-free model matching.

Song Wang, Li Sun, Wei Fan et al.

Combining deep learning and geometric modeling, the obstacle detection system improves by 50% on the Lost and Found dataset.

Sebastian Ramos, Stefan Gehrig, Peter Pinggera et al.

This method employs a convnet trained solely on static images, guided by previous frame masks, achieving competitive video object segmentation performance.

Anna Khoreva, Federico Perazzi, Rodrigo Benenson et al.

DeMoN learns depth and camera motion from monocular image pairs, achieving higher accuracy and robustness.

Benjamin Ummenhofer, Huizhong Zhou, Jonas Uhrig et al.

FlowNet2 combines curriculum training, warped stacking, and small-motion specialization, reaching 3.96/6.02 EPE on Sintel test at about 123 ms.

Eddy Ilg, Nikolaus Mayer, Tonmoy Saikia et al.

Proposes an FCN-LSTM-based end-to-end driving model trained on large-scale crowd-sourced videos, achieving 84.1% action prediction accuracy.

Huazhe Xu, Yang Gao, Fisher Yu et al.

Constructed a balanced VQA dataset with similar image pairs to reduce language bias, improving visual understanding and interpretability.

Yash Goyal, Tejas Khot, Douglas Summers-Stay et al.

SyncSpecCNN uses spectral domain parameterization and spectral transformer networks to achieve state-of-the-art 3D shape segmentation and keypoint prediction, excelling on benchmarks.

Li Yi, Hao Su, Xingwen Guo et al.