In recent years, large image data sets such as "Ima-geNet", "TinyImages" or ever-growing social networks like "Flickr" have emerged, posing new challenges to image classification that were not apparent in smaller image sets. In particular, the efficient handling of dynamically growing data sets, where not only the amount of training images, but also the number of classes increases over time, is a relatively unexplored problem. To remedy this, we introduce Nearest Class Mean Forests (NCMF), a variant of Random Forests where the decision nodes are based on nearest class mean (NCM) classification. NCMFs not only outperform conventional random forests, but are also well suited for integrating new classes. To this end, we propose and compare several approaches to incorporate data from new classes, so as to seamlessly extend the previously trained forest instead of re-training them from scratch. In our experiments, we show that NCMFs trained on small data sets with 10 classes can be extended to large data sets with 1000 classes without significant loss of accuracy compared to training from scratch on the full data.

Large image datasets such as ImageNet or open-ended photo websites like Flickr are revealing new challenges to image classification that were not apparent in smaller, fixed sets. In particular, the efficient handling of dynamically growing datasets, where not only the amount of training data but also the number of classes increases over time, is a relatively unexplored problem. In this challenging setting, we study how two variants of Random Forests (RF) perform under four strategies to incorporate new classes while avoiding to retrain the RFs from scratch. The various strategies account for different trade-offs between classification accuracy and computational efficiency. In our extensive experiments, we show that both RF variants, one based on Nearest Class Mean classifiers and the other on SVMs, outperform conventional RFs and are well suited for incrementally learning new classes. In particular, we show that RFs initially trained with just 10 classes can be extended to 1,000 classes with an acceptable loss of accuracy compared to training from the full data and with great computational savings compared to retraining for each new batch of classes.

The number of digital images is growing extremely rapidly, and so is the need for their classification. But, as more images of pre-defined categories become available, they also become more diverse and cover finer semantic differences. Ultimately, the categories themselves need to be divided into subcategories to account for that semantic refinement. Image classification in general has improved significantly over the last few years, but it still requires a massive amount of manually annotated data. Subdividing categories into subcategories multiples the number of labels, aggravating the annotation problem. Hence, we can expect the annotations to be refined only for a subset of the already labeled data, and exploit coarser labeled data to improve classification. In this work, we investigate how coarse category labels can be used to improve the classification of subcategories. To this end, we adopt the framework of Random Forests and propose a regularized objective function that takes into account relations between categories and subcategories. Compared to approaches that disregard the extra coarse labeled data, we achieve a relative improvement in subcategory classification accuracy of up to 22% in our large-scale image classification experiments.

State-of-the-art methods for object detection are mostly based on an expensive exhaustive search over the image at different scales. In order to reduce the computational time, one can perform a selective search to obtain a small subset of relevant object hypotheses that need to be evaluated by the detector. For that purpose, we employ a regression to predict possible object scales and locations by exploiting the local context of an image. Furthermore, we show how a priori information, if available, can be integrated to improve the prediction. The experimental results on three datasets including the Caltech pedestrian and PASCAL VOC dataset show that our method achieves the detection performance of an exhaustive search approach with much less computational load. Since we model the prior distribution over the proposals locally, it generalizes well and can be successfully applied across datasets.

Fashion is a major segment in e-commerce with growing importance and a steadily increasing number of products. Since manual annotation of apparel items is very tedious, the product databases need to be organized automatically, e.g. by image classification. Common image classification approaches are based on features engineered for general purposes which perform poorly on specific images of apparel. We therefore propose to learn discriminative features based on a small set of annotated images. We experimentally evaluate our method on a dataset with 30,000 images containing apparel items, and compare it to other engineered and learned sets of features. The classification accuracy of our features is significantly superior to designed HOG and SIFT features (43.7% and 16.1% relative improvement, respectively). Our method allows for fast feature extraction and training, is easy to implement and, unlike deep convolutional networks, does not require powerful dedicated hardware.