Radu Timofte

Pedestrian detection at 100 frames per second

We present a new pedestrian detector that improves both in speed and quality over state-of-the-art. By efficiently handling different scales and transferring computation from test time to training time, detection speed is improved. When processing monocular images, our system provides high quality detections at 50 fps. We also propose a new method for exploiting geometric context extracted from stereo images. On a single CPU+GPU desktop machine, we reach 135 fps, when processing street scenes, from rectified input to detections output.

Anchored Neighborhood Regression for Fast Example-Based Super-Resolution

Recently there have been significant advances in image up scaling or image super-resolution based on a dictionary of low and high resolution exemplars. The running time of the methods is often ignored despite the fact that it is a critical factor for real applications. This paper proposes fast super-resolution methods while making no compromise on quality. First, we support the use of sparse learned dictionaries in combination with neighbor embedding methods. In this case, the nearest neighbors are computed using the correlation with the dictionary atoms rather than the Euclidean distance. Moreover, we show that most of the current approaches reach top performance for the right parameters. Second, we show that using global collaborative coding has considerable speed advantages, reducing the super-resolution mapping to a precomputed projective matrix. Third, we propose the anchored neighborhood regression. That is to anchor the neighborhood embedding of a low resolution patch to the nearest atom in the dictionary and to precompute the corresponding embedding matrix. These proposals are contrasted with current state-of-the-art methods on standard images. We obtain similar or improved quality and one or two orders of magnitude speed improvements.

A+: Adjusted Anchored Neighborhood Regression for Fast Super-Resolution

We address the problem of image upscaling in the form of single image super-resolution based on a dictionary of low- and high-resolution exemplars. Two recently proposed methods, Anchored Neighborhood Regression (ANR) and Simple Functions (SF), provide state-of-the-art quality performance. Moreover, ANR is among the fastest known super-resolution methods. ANR learns sparse dictionaries and regressors anchored to the dictionary atoms. SF relies on clusters and corresponding learned functions. We propose A+, an improved variant of ANR, which combines the best qualities of ANR and SF. A+ builds on the features and anchored regressors from ANR but instead of learning the regressors on the dictionary it uses the full training material, similar to SF. We validate our method on standard images and compare with state-of-the-art methods. We obtain improved quality (i.e. 0.2–0.7 dB PSNR better than ANR) and excellent time complexity, rendering A+ the most efficient dictionary-based super-resolution method to date.

Hough transform and 3D SURF for robust three dimensional classification

Most methods for the recognition of shape classes from 3D datasets focus on classifying clean, often manually generated models. However, 3D shapes obtained through acquisition techniques such as Structure-from-Motion or LIDAR scanning are noisy, clutter and holes. In that case global shape features--still dominating the 3D shape class recognition literature--are less appropriate. Inspired by 2D methods, recently researchers have started to work with local features. In keeping with this strand, we propose a new robust 3D shape classification method. It contains two main contributions. First, we extend a robust 2D feature descriptor, SURF, to be used in the context of 3D shapes. Second, we show how 3D shape class recognition can be improved by probabilistic Hough transform based methods, already popular in 2D. Through our experiments on partial shape retrieval, we show the power of the proposed 3D features. Their combination with the Hough transform yields superior results for class recognition on standard datasets. The potential for the applicability of such a method in classifying 3D obtained from Structure-from-Motion methods is promising, as we show in some initial experiments.

Traffic sign recognition — How far are we from the solution?

Traffic sign recognition has been a recurring application domain for visual objects detection. The public datasets have only recently reached large enough size and variety to enable proper empirical studies. We revisit the topic by showing how modern methods perform on two large detection and classification datasets (thousand of images, tens of categories) captured in Belgium and Germany. We show that, without any application specific modification, existing methods for pedestrian detection, and for digit and face classification; can reach performances in the range of 95% ~ 99% of the perfect solution. We show detailed experiments and discuss the trade-off of different options. Our top performing methods use modern variants of HOG features for detection, and sparse representations for classification.

Multi-view traffic sign detection, recognition, and 3D localisation

Several applications require information about street furniture. Part of the task is to survey all traffic signs. This has to be done for millions of km of road, and the exercise needs to be repeated every so often. We used a van with eight roof-mounted cameras to drive through the streets and took images every meter. The paper proposes a pipeline for the efficient detection and recognition of traffic signs from such images. The task is challenging, as illumination conditions change regularly, occlusions are frequent, sign positions and orientations vary substantially, and the actual signs are far less similar among equal types than one might expect. We combine 2D and 3D techniques to improve results beyond the state-of-the-art, which is still very much preoccupied with single view analysis. For the initial detection in single frames, we use a set of colour- and shape-based criteria. They yield a set of candidate sign patterns. The selection of such candidates allows for a significant speed up over a sliding window approach while keeping similar performance. A speedup is also achieved through a proposed efficient bounded evaluation of AdaBoost detectors. The 2D detections in multiple views are subsequently combined to generate 3D hypotheses. A Minimum Description Length formulation yields the set of 3D traffic signs that best explains the 2D detections. The paper comes with a publicly available database, with more than 13,000 traffic signs annotations.

DEX: Deep EXpectation of Apparent Age from a Single Image

In this paper we tackle the estimation of apparent age in still face images with deep learning. Our convolutional neural networks (CNNs) use the VGG-16 architecture [13] and are pretrained on ImageNet for image classification. In addition, due to the limited number of apparent age annotated images, we explore the benefit of finetuning over crawled Internet face images with available age. We crawled 0.5 million images of celebrities from IMDB and Wikipedia that we make public. This is the largest public dataset for age prediction to date. We pose the age regression problem as a deep classification problem followed by a softmax expected value refinement and show improvements over direct regression training of CNNs. Our proposed method, Deep EXpectation (DEX) of apparent age, first detects the face in the test image and then extracts the CNN predictions from an ensemble of 20 networks on the cropped face. The CNNs of DEX were finetuned on the crawled images and then on the provided images with apparent age annotations. DEX does not use explicit facial landmarks. Our DEX is the winner (1st place) of the ChaLearn LAP 2015 challenge on apparent age estimation with 115 registered teams, significantly outperforming the human reference.

Deep Expectation of Real and Apparent Age from a Single Image Without Facial Landmarks

In this paper we propose a deep learning solution to age estimation from a single face image without the use of facial landmarks and introduce the IMDB-WIKI dataset, the largest public dataset of face images with age and gender labels. If the real age estimation research spans over decades, the study of apparent age estimation or the age as perceived by other humans from a face image is a recent endeavor. We tackle both tasks with our convolutional neural networks (CNNs) of VGG-16 architecture which are pre-trained on ImageNet for image classification. We pose the age estimation problem as a deep classification problem followed by a softmax expected value refinement. The key factors of our solution are: deep learned models from large data, robust face alignment, and expected value formulation for age regression. We validate our methods on standard benchmarks and achieve state-of-the-art results for both real and apparent age estimation.

Jointly Optimized Regressors for Image Super-resolution

Learning regressors from low‐resolution patches to high‐resolution patches has shown promising results for image super‐resolution. We observe that some regressors are better at dealing with certain cases, and others with different cases. In this paper, we jointly learn a collection of regressors, which collectively yield the smallest super‐resolving error for all training data. After training, each training sample is associated with a label to indicate its ‘best’ regressor, the one yielding the smallest error. During testing, our method bases on the concept of ‘adaptive selection’ to select the most appropriate regressor for each input patch. We assume that similar patches can be super‐resolved by the same regressor and use a fast, approximate kNN approach to transfer the labels of training patches to test patches. The method is conceptually simple and computationally efficient, yet very effective. Experiments on four datasets show that our method outperforms competing methods.

Handling Occlusions with Franken-Classifiers

Detecting partially occluded pedestrians is challenging. A common practice to maximize detection quality is to train a set of occlusion-specific classifiers, each for a certain amount and type of occlusion. Since training classifiers is expensive, only a handful are typically trained. We show that by using many occlusion-specific classifiers, we outperform previous approaches on three pedestrian datasets, INRIA, ETH, and Caltech USA. We present a new approach to train such classifiers. By reusing computations among different training stages, 16 occlusion-specific classifiers can be trained at only one tenth the cost of one full training. We show that also test time cost grows sub-linearly.