Incremental learning of object detectors using a visual shape alphabet

Feature context for image classification and object detection

“…Under the RP AUC measure, we get higher results than [22] in 5 of the 7 classes. Under the RP EER measure, we also get higher results than [18] in 5 of the 7 classes.…”

“…FC is also suitable to describe objects when only the bounding boxes of objects are given in training. Using sliding window strategy, FC can outperform more sophisticated object detectors [22,18,21], which is a nontrivial accomplishment. In the future, we would like to combine FC method with the interest region detectors.…”

“…As show in Table 3, we use the same number of training and testing images with [22] and [18]. Under the RP AUC measure, we get higher results than [22] in 5 of the 7 classes.…”

“…We plot recall-precision curve for each class, then calculate the area under the curve (AUC) and equal error rate (EER) to compare with two state-of-the-art methods [22] and [18]. As show in Table 3, we use the same number of training and testing images with [22] and [18].…”

3D Object Detection and Viewpoint Selection in Sketch Images Using Local Patch-Based Zernike Moments

“…We chose to tackle the problem by recognizing a 3D object through a set of 2D images each corresponding to a single viewpoint. In the state of the art, we can find a great variety of 2D object recognition methods, some successful methods are those of Lowe [11], Belongie et al [2], Furgus et al [5], Shotton et al [14], Opelt et al [12]. However, most of these methods work only on textured objects and fail on smooth shapes, such as sketch images (drawings).…”