Reducing bit-widths of activations and weights of deep networks makes it efficient to compute and store them in memory, which is crucial in their deployments to resourcelimited devices, such as mobile phones. However, decreasing bit-widths with quantization generally yields drastically degraded accuracy. To tackle this problem, we propose to learn to quantize activations and weights via a trainable quantizer that transforms and discretizes them. Specifically, we parameterize the quantization intervals and obtain their optimal values by directly minimizing the task loss of the network. This quantization-interval-learning (QIL) allows the quantized networks to maintain the accuracy of the fullprecision (32-bit) networks with bit-width as low as 4-bit and minimize the accuracy degeneration with further bitwidth reduction (i.e., 3 and 2-bit). Moreover, our quantizer can be trained on a heterogeneous dataset, and thus can be used to quantize pretrained networks without access to their training data. We demonstrate the effectiveness of our trainable quantizer on ImageNet dataset with various network architectures such as ResNet-18, -34 and AlexNet, on which it outperforms existing methods to achieve the stateof-the-art accuracy.

In this paper, we describe a technique to automatically enhance the perceptual quality of an image. Unlike previous techniques, where global statistics of the image are used to determine enhancement operation, our method is local and relies on local scene descriptors and context in addition to high-level image statistics. We cast the problem of image enhancement as searching for the best transformation for each pixel in the given image and then discovering the enhanced image using a formulation based on Gaussian Random Fields. The search is done in a coarse-to-fine manner, namely by finding the best candidate images, followed by pixels. Our experiments indicate that such context-based local enhancement is better than global enhancement schemes. A user study using Mechanical Turk shows that the subjects prefer contextual and local enhancements over the ones provided by existing schemes.

No abstract

We introduce an approach to image retrieval and auto-tagging that leverages the implicit information about object importance conveyed by the list of keyword tags a person supplies for an image. We propose an unsupervised learning procedure based on Kernel Canonical Correlation Analysis that discovers the relationship between how humans tag images (e.g., the order in which words are mentioned) and the relative importance of objects and their layout in the scene. Using this discovered connection, we show how to boost accuracy for novel queries, such that the search results better preserve the aspects a human may find most worth mentioning. We evaluate our approach on three datasets using either keyword tags or natural language descriptions, and quantify results with both ground truth parameters as well as direct tests with human subjects. Our results show clear improvements over approaches that either rely on image features alone, or that use words and image features but ignore the implied importance cues. Overall, our work provides a novel way to incorporate high-level human perception of scenes into visual representations for enhanced image search.

No abstract