DenseNet introduces dense connectivity, connecting each layer to all previous layers, reducing parameters and improving gradient flow; achieves state-of-the-art results on CIFAR-10/100, SVHN, ImageNet.

Gao Huang, Zhuang Liu, Laurens van der Maaten et al.

Proposes ECTC framework for weakly supervised video action labeling, leveraging frame-to-frame visual similarity to improve alignment accuracy.

De-An Huang, Li Fei-Fei, Juan Carlos Niebles

Using FGSM and iterative attacks, physical adversarial examples remain effective; over 50% cause misclassification after camera capture.

Alexey Kurakin, Ian Goodfellow, Samy Bengio

Li and Hoiem propose Learning without Forgetting (LwF), a method that preserves old task performance using only new task data via knowledge distillation, without access to old data.

Zhizhong Li, Derek Hoiem

Conditional PixelCNN generates images, improving PixelRNN performance on ImageNet.

Aaron van den Oord, Nal Kalchbrenner, Oriol Vinyals et al.

V-Net uses a Dice-coefficient-based loss in a 3D fully convolutional network for prostate MRI segmentation, achieving high accuracy and speed.

Fausto Milletari, Nassir Navab, Seyed-Ahmad Ahmadi

Proposed a regularization method using stochastic transformations and perturbations for deep semi-supervised learning, improving model accuracy.

Mehdi Sajjadi, Mehran Javanmardi, Tolga Tasdizen

Proposes convolutional neural fabrics embedding exponential architectures, sharing parameters, achieving competitive results on MNIST, CIFAR10, and Part Labels datasets.

Shreyas Saxena, Jakob Verbeek

Fully convolutional residual network achieves state-of-the-art monocular depth prediction with fewer parameters, less training data, and real-time inference.

Iro Laina, Christian Rupprecht, Vasileios Belagiannis et al.

A bidirectional LSTM with DPP for video summarization, achieving state-of-the-art F-score of 42.9% on SumMe and 59.6% on TVSum.

Ke Zhang, Wei-Lun Chao, Fei Sha et al.

Proposes FractalNet, a deep neural architecture based on self-similarity without residuals, matching ResNet performance with drop-path regularization.

Gustav Larsson, Michael Maire, Gregory Shakhnarovich

Proposed a category-level semantic segmentation-based instance segmentation method, achieving 79.1% mIoU on PASCAL VOC 2012, surpassing detect-then-segment approaches.

Zifeng Wu, Chunhua Shen, Anton van den Hengel

R-FCN introduces position-sensitive score maps in a fully convolutional framework, achieving 83.6% mAP on VOC 2007 with 2.5× speedup over Faster R-CNN.

Jifeng Dai, Yi Li, Kaiming He et al.

Proposes end-to-end training of neural language models for fine-grained zero-shot visual recognition, outperforming attribute-based methods.

Scott Reed, Zeynep Akata, Bernt Schiele et al.

Proposes a deep network with relative depth annotations and a new wild scene dataset, significantly improving single-image depth estimation.

Weifeng Chen, Zhao Fu, Dawei Yang et al.

Deep3D employs end-to-end deep CNNs to convert 2D videos into high-quality 3D stereo pairs, outperforming traditional methods with MAE of 6.87 on movie datasets.

Junyuan Xie, Ross Girshick, Ali Farhadi

Proposed 3D-R2N2 uses spatially-structured 3D-LSTM for multi-view 3D reconstruction without labels, outperforming state-of-the-art.

Christopher B. Choy, Danfei Xu, JunYoung Gwak et al.

Unsupervised learning of visual representations by solving jigsaw puzzles, outperforming existing methods.

Mehdi Noroozi, Paolo Favaro

Proposes InstanceFCN, using pixel-relative position classification for instance proposals, achieving AR@100 of 49.7% on Pascal VOC.

Jifeng Dai, Kaiming He, Yi Li et al.

Combining Gaussian CRFs with deep learning for semantic segmentation, achieving significant improvements on VOC PASCAL 2012.

Siddhartha Chandra, Iasonas Kokkinos