Aditya Khosla

Learning Deep Features for Discriminative Localization

In this work, we revisit the global average pooling layer proposed in [13], and shed light on how it explicitly enables the convolutional neural network (CNN) to have remarkable localization ability despite being trained on imagelevel labels. While this technique was previously proposed as a means for regularizing training, we find that it actually builds a generic localizable deep representation that exposes the implicit attention of CNNs on an image. Despite the apparent simplicity of global average pooling, we are able to achieve 37.1% top-5 error for object localization on ILSVRC 2014 without training on any bounding box annotation. We demonstrate in a variety of experiments that our network is able to localize the discriminative image regions despite just being trained for solving classification task1.

3D ShapeNets: A deep representation for volumetric shapes

3D shape is a crucial but heavily underutilized cue in todays computer vision systems, mostly due to the lack of a good generic shape representation. With the recent availability of inexpensive 2.5D depth sensors (e.g. Microsoft Kinect), it is becoming increasingly important to have a powerful 3D shape representation in the loop. Apart from category recognition, recovering full 3D shapes from view-based 2.5D depth maps is also a critical part of visual understanding. To this end, we propose to represent a geometric 3D shape as a probability distribution of binary variables on a 3D voxel grid, using a Convolutional Deep Belief Network. Our model, 3D ShapeNets, learns the distribution of complex 3D shapes across different object categories and arbitrary poses from raw CAD data, and discovers hierarchical compositional part representation automatically. It naturally supports joint object recognition and shape completion from 2.5D depth maps, and it enables active object recognition through view planning. To train our 3D deep learning model, we construct ModelNet - a large-scale 3D CAD model dataset. Extensive experiments show that our 3D deep representation enables significant performance improvement over the-state-of-the-arts in a variety of tasks.

Human action recognition by learning bases of action attributes and parts

In this work, we propose to use attributes and parts for recognizing human actions in still images. We define action attributes as the verbs that describe the properties of human actions, while the parts of actions are objects and poselets that are closely related to the actions. We jointly model the attributes and parts by learning a set of sparse bases that are shown to carry much semantic meaning. Then, the attributes and parts of an action image can be reconstructed from sparse coefficients with respect to the learned bases. This dual sparsity provides theoretical guarantee of our bases learning and feature reconstruction approach. On the PASCAL action dataset and a new “Stanford 40 Actions” dataset, we show that our method extracts meaningful high-order interactions between attributes and parts in human actions while achieving state-of-the-art classification performance.

Combining randomization and discrimination for fine-grained image categorization

In this paper, we study the problem of fine-grained image categorization. The goal of our method is to explore fine image statistics and identify the discriminative image patches for recognition. We achieve this goal by combining two ideas, discriminative feature mining and randomization. Discriminative feature mining allows us to model the detailed information that distinguishes different classes of images, while randomization allows us to handle the huge feature space and prevents over-fitting. We propose a random forest with discriminative decision trees algorithm, where every tree node is a discriminative classifier that is trained by combining the information in this node as well as all upstream nodes. Our method is tested on both subordinate categorization and activity recognition datasets. Experimental results show that our method identifies semantically meaningful visual information and outperforms state-of-the-art algorithms on various datasets.

Undoing the damage of dataset bias

The presence of bias in existing object recognition datasets is now well-known in the computer vision community. While it remains in question whether creating an unbiased dataset is possible given limited resources, in this work we propose a discriminative framework that directly exploits dataset bias during training. In particular, our model learns two sets of weights: (1) bias vectors associated with each individual dataset, and (2) visual world weights that are common to all datasets, which are learned by undoing the associated bias from each dataset. The visual world weights are expected to be our best possible approximation to the object model trained on an unbiased dataset, and thus tend to have good generalization ability. We demonstrate the effectiveness of our model by applying the learned weights to a novel, unseen dataset, and report superior results for both classification and detection tasks compared to a classical SVM that does not account for the presence of bias. Overall, we find that it is beneficial to explicitly account for bias when combining multiple datasets.

What makes an image popular

Hundreds of thousands of photographs are uploaded to the internet every minute through various social networking and photo sharing platforms. While some images get millions of views, others are completely ignored. Even from the same users, different photographs receive different number of views. This begs the question: What makes a photograph popular? Can we predict the number of views a photograph will receive even before it is uploaded? These are some of the questions we address in this work. We investigate two key components of an image that affect its popularity, namely the image content and social context. Using a dataset of about 2.3 million images from Flickr, we demonstrate that we can reliably predict the normalized view count of images with a rank correlation of 0.81 using both image content and social cues. In this paper, we show the importance of image cues such as color, gradients, deep learning features and the set of objects present, as well as the importance of various social cues such as number of friends or number of photos uploaded that lead to high or low popularity of images.

Large-Scale Video Summarization Using Web-Image Priors

Given the enormous growth in user-generated videos, it is becoming increasingly important to be able to navigate them efficiently. As these videos are generally of poor quality, summarization methods designed for well-produced videos do not generalize to them. To address this challenge, we propose to use web-images as a prior to facilitate summarization of user-generated videos. Our main intuition is that people tend to take pictures of objects to capture them in a maximally informative way. Such images could therefore be used as prior information to summarize videos containing a similar set of objects. In this work, we apply our novel insight to develop a summarization algorithm that uses the web-image based prior information in an unsupervised manner. Moreover, to automatically evaluate summarization algorithms on a large scale, we propose a framework that relies on multiple summaries obtained through crowdsourcing. We demonstrate the effectiveness of our evaluation framework by comparing its performance to that of multiple human evaluators. Finally, we present results for our framework tested on hundreds of user-generated videos.

Places: A 10 Million Image Database for Scene Recognition

The rise of multi-million-item dataset initiatives has enabled data-hungry machine learning algorithms to reach near-human semantic classification performance at tasks such as visual object and scene recognition. Here we describe the Places Database, a repository of 10 million scene photographs, labeled with scene semantic categories, comprising a large and diverse list of the types of environments encountered in the world. Using the state-of-the-art Convolutional Neural Networks (CNNs), we provide scene classification CNNs (Places-CNNs) as baselines, that significantly outperform the previous approaches. Visualization of the CNNs trained on Places shows that object detectors emerge as an intermediate representation of scene classification. With its high-coverage and high-diversity of exemplars, the Places Database along with the Places-CNNs offer a novel resource to guide future progress on scene recognition problems.

Network Dissection: Quantifying Interpretability of Deep Visual Representations

We propose a general framework called Network Dissection for quantifying the interpretability of latent representations of CNNs by evaluating the alignment between individual hidden units and a set of semantic concepts. Given any CNN model, the proposed method draws on a data set of concepts to score the semantics of hidden units at each intermediate convolutional layer. The units with semantics are labeled across a broad range of visual concepts including objects, parts, scenes, textures, materials, and colors. We use the proposed method to test the hypothesis that interpretability is an axis-independent property of the representation space, then we apply the method to compare the latent representations of various networks when trained to solve different classification problems. We further analyze the effect of training iterations, compare networks trained with different initializations, and measure the effect of dropout and batch normalization on the interpretability of deep visual representations. We demonstrate that the proposed method can shed light on characteristics of CNN models and training methods that go beyond measurements of their discriminative power.

Comparison of deep neural networks to spatio-temporal cortical dynamics of human visual object recognition reveals hierarchical correspondence

The complex multi-stage architecture of cortical visual pathways provides the neural basis for efficient visual object recognition in humans. However, the stage-wise computations therein remain poorly understood. Here, we compared temporal (magnetoencephalography) and spatial (functional MRI) visual brain representations with representations in an artificial deep neural network (DNN) tuned to the statistics of real-world visual recognition. We showed that the DNN captured the stages of human visual processing in both time and space from early visual areas towards the dorsal and ventral streams. Further investigation of crucial DNN parameters revealed that while model architecture was important, training on real-world categorization was necessary to enforce spatio-temporal hierarchical relationships with the brain. Together our results provide an algorithmically informed view on the spatio-temporal dynamics of visual object recognition in the human visual brain.