Learning to Detect Small Signs from Large Images – Automated localization systems are among the most important tools for recognizing image objects in video. Recent work has demonstrated that machine-generated images can be used to train a classifier of object detection methods. In this work, we are interested in learning to associate the features of a object to its position, which we also refer to as the camera position. We exploit a deep recurrent network for image training that learns this joint representation using the input features of the network for this purpose. Experiments on the MNIST dataset show that the proposed method outperforms the state of the art methods in several image detection tasks.

We present a novel multi-view feature representation learning method for automatic segmentation of facial landmarks in images. We show that the proposed algorithm outperforms baseline approaches, with significant improvement of performance compared to the traditional approach. Additionally, we present a new benchmark dataset for automatically segmenting landmarks in images at human and machine levels using multi-view convolutional neural networks. Extensive evaluation on two standard benchmark datasets for facial landmarks segmentation shows that our framework significantly outperforms baseline approaches.

Learning Disentangled Representations with Latent Factor Modeling

Efficient Ranking of High Dimensional Features using a random partition function, using a generalized kernel

Learning to Detect Small Signs from Large Images

Learning an Integrated Deep Filter based on Hybrid Coherent Cuts

Multi-View Representation Lasso through Constrained Random Projections for Image RecognitionWe present a novel multi-view feature representation learning method for automatic segmentation of facial landmarks in images. We show that the proposed algorithm outperforms baseline approaches, with significant improvement of performance compared to the traditional approach. Additionally, we present a new benchmark dataset for automatically segmenting landmarks in images at human and machine levels using multi-view convolutional neural networks. Extensive evaluation on two standard benchmark datasets for facial landmarks segmentation shows that our framework significantly outperforms baseline approaches.