Proposes Gated-SCNN, a dual-stream network with gating mechanisms, boosting city scene segmentation by 2% mIoU and 4% boundary F-score, especially improving small and thin objects.

Towaki Takikawa, David Acuna, Varun Jampani et al.

Proposes a multi-dataset mixing approach with scale and shift-invariant loss for robust monocular depth estimation, achieving superior zero-shot cross-dataset transfer on five diverse datasets.

René Ranftl, Katrin Lasinger, David Hafner et al.

LV-Net combines two-stream pyramid and nested encoder-decoder for robust salient object detection in optical RSIs, outperforming 14 SOTA methods on the ORSSD dataset.

Chongyi Li, Runmin Cong, Junhui Hou et al.

This study presents RGBD multimodal end-to-end autonomous driving, with early fusion outperforming single-modality models.

Yi Xiao, Felipe Codevilla, Akhil Gurram et al.

Proposes Scene Representation Networks (SRNs), a continuous, differentiable scene model trained end-to-end from 2D images, capturing geometry and appearance without supervision.

Vincent Sitzmann, Michael Zollhöfer, Gordon Wetzstein

3D-BoNet directly predicts 3D bounding boxes and instance masks on point clouds, achieving 10x efficiency gains and surpassing ScanNet and S3DIS benchmarks.

Bo Yang, Jianan Wang, Ronald Clark et al.

Proposes Patch-wise Adversarial Regularization (PAR) to improve CNN robustness by penalizing local predictive power, enhancing out-of-domain generalization.

Haohan Wang, Songwei Ge, Eric P. Xing et al.

Variational Autoencoder-based 6-DOF grasp generation achieves 88% success, enabling seamless simulation-to-real transfer.

Arsalan Mousavian, Clemens Eppner, Dieter Fox

Proposed a 'loss prediction module' for active learning, achieving a 0.42% accuracy increase on CIFAR-10.

Donggeun Yoo, In So Kweon

S4L combines self-supervised and semi-supervised learning, achieving new SOTA on ILSVRC-2012 with only 10% labels.

Xiaohua Zhai, Avital Oliver, Alexander Kolesnikov et al.

FANTrack employs deep CNNs for 3D multi-object data association, achieving 77.72% MOTA on KITTI.

Erkan Baser, Venkateshwaran Balasubramanian, Prarthana Bhattacharyya et al.

Scaling Jigsaw and Colorization to 100M images reveals self-supervised learning's potential to rival supervised methods in specific tasks.

Priya Goyal, Dhruv Mahajan, Abhinav Gupta et al.

DRIT++ achieves diverse image-to-image translation via disentangled representations, evaluated using Fréchet Inception Distance and perceptual distance.

Hsin-Ying Lee, Hung-Yu Tseng, Qi Mao et al.

Proposed a billion-scale semi-supervised learning method using teacher/student paradigm; ResNet-50 achieved 81.2% accuracy on ImageNet.

I. Zeki Yalniz, Hervé Jégou, Kan Chen et al.

HAWQ leverages Hessian spectrum for automatic mixed-precision quantization, achieving 8× activation compression on ResNet20 with only 0.15% accuracy loss.

Zhen Dong, Zhewei Yao, Amir Gholami et al.

This paper introduces a free-form video inpainting model using 3D Gated Convolution and Temporal PatchGAN, significantly improving spatial detail and temporal consistency.

Ya-Liang Chang, Zhe Yu Liu, Kuan-Ying Lee et al.

Social Ways uses Info-GAN to preserve multimodal pedestrian futures, reaching 0.39/0.64 ADE/FDE on ETH.

Javad Amirian, Jean-Bernard Hayet, Julien Pettre

Behavior cloning achieves state-of-the-art in complex driving but faces generalization and data bias issues.

Felipe Codevilla, Eder Santana, Antonio M. López et al.

KPConv introduces flexible, deformable point cloud convolution, achieving 92.9% accuracy on ModelNet40.

Hugues Thomas, Charles R. Qi, Jean-Emmanuel Deschaud et al.

Proposes MinkowskiNet, a 4D sparse convolutional neural network leveraging generalized sparse convolutions, achieving 67.9% mIoU on ScanNet, outperforming prior methods by 19%.

Christopher Choy, JunYoung Gwak, Silvio Savarese