• Publications
  • Influence
Weakly Supervised Object Localization with Multi-Fold Multiple Instance Learning
TLDR
We propose a weakly supervised learning method to learn object detectors from image-wide labels, which improves the localization accuracy by incorporating an objectness prior. Expand
  • 318
  • 41
  • PDF
Semantic Segmentation using Adversarial Networks
TLDR
We present, to the best of our knowledge, the first application of adversarial training to semantic segmentation. Expand
  • 445
  • 30
  • PDF
FeaStNet: Feature-Steered Graph Convolutions for 3D Shape Analysis
TLDR
We propose a novel graph-convolution operator to establish correspondences between filter weights and graph neighborhoods with arbitrary connectivity. Expand
  • 149
  • 17
  • PDF
Heterogeneous Face Recognition with CNNs
TLDR
Heterogeneous face recognition aims to recognize faces across different sensor modalities. Expand
  • 68
  • 11
  • PDF
Predicting Future Instance Segmentations by Forecasting Convolutional Features
TLDR
We develop a predictive model in the space of fixed-sized convolutional features of the Mask R-CNN instance segmentation model. Expand
  • 43
  • 10
  • PDF
Convolutional Neural Fabrics
TLDR
We propose a "fabric" that embeds an exponentially large number of architectures. Expand
  • 156
  • 6
  • PDF
The LEAR submission at Thumos 2014
TLDR
We describe the submission of the INRIA LEAR team to the THU-MOS workshop in conjunction with ECCV 2014. Expand
  • 114
  • 5
  • PDF
Pervasive Attention: 2D Convolutional Neural Networks for Sequence-to-Sequence Prediction
TLDR
We propose an alternative neural MT architecture which instead relies on a single 2D convolutional neural network across both sequences, while being conceptually simpler and having fewer parameters. Expand
  • 47
  • 4
  • PDF
Scene Coordinate Regression with Angle-Based Reprojection Loss for Camera Relocalization
TLDR
In this paper, we present a new angle-based reprojection loss, which resolves the issues of the original reproject loss, and the system achieves more accurate results. Expand
  • 18
  • 4
  • PDF
Frankenstein: Learning Deep Face Representations Using Small Data
TLDR
We propose a method to generate very large training data sets of synthetic images by compositing real face images in a given data set. Expand
  • 72
  • 3
  • PDF