Browsing by Author "Vidal, R."

Now showing 1 - 8 of 8
  • Thumbnail Image
    Item
    CompactNets: Compact Hierarchical Compositional Networks for Visual Recognition
    (2020) Löbel Díaz, Hans-Albert; Vidal, R.; Soto Arriaza, Álvaro Marcelo
  • Thumbnail Image
    Item
    Functional role of the disulfide isomerase ERp57 in axonal regeneration
    (2015) Castillo, V.; Onate, M.; Woehlbier, U.; Rozas, P.; Andreu, C.; Medinas, D.; Valdes, P.; Osorio, F.; Mercado, G.; Court G., Felipe; Vidal, R.; Kerr, B.; Hetz, C.
  • No Thumbnail Available
    Item
    Joint dictionary and classifier learning for categorization of images using a max-margin framework
    (2014) Lobel, H.; Vidal, R.; Mery Quiroz, Domingo Arturo; Soto, A.
  • Thumbnail Image
    Item
    Learning Shared, Discriminative, and Compact Representations for Visual Recognition
    (2015) Löbel Díaz, Hans-Albert; Vidal, R.; Soto Arriaza, Álvaro Marcelo
  • No Thumbnail Available
    Item
    Learning Shared, Discriminative, and Compact Representations for Visual Recognition
    (IEEE, 2015) Löbel Díaz, Hans-Albert; Vidal, R.; Soto Arriaza, Álvaro Marcelo
    Dictionary-based and part-based methods are among the most popular approaches to visual recognition. In both methods, a mid-level representation is built on top of low-level image descriptors and high-level classifiers are trained on top of the mid-level representation. While earlier methods built the mid-level representation without supervision, there is currently great interest in learning both representations jointly to make the mid-level representation more discriminative. In this work we propose a new approach to visual recognition that jointly learns a shared, discriminative, and compact mid-level representation and a compact high-level representation. By using a structured output learning framework, our approach directly handles the multiclass case at both levels of abstraction. Moreover, by using a group-sparse prior in the structured output learning framework, our approach encourages sharing of visual words and thus reduces the number of words used to represent each class. We test our proposed method on several popular benchmarks. Our results show that, by jointly learning midand high-level representations, and fostering the sharing of discriminative visual words among target classes, we are able to achieve state-of-the-art recognition performance using far less visual words than previous approaches.
  • Thumbnail Image
    Item
    Regulation of Memory Formation by the Transcription Factor XBP1
    (2016) Martínez, Gonzalo; Vidal, R.; Mardones, P.; Serrano, F.; Ardiles, A.; Wirth, C.; Valdés, P.; Thielen, P.; Schneider, B.; Inestrosa Cantín, Nibaldo; Kerr, B.; Valdés, J.; Palacios, A
  • Thumbnail Image
    Item
    Segmentation of circular casting defects using a robust algorithm
    (BRITISH INST NON-DESTRUCTIVE TESTING, 2005) Ghoreyshi, A.; Vidal, R.; Mery, D.
    In this paper, we describe three methods for detecting defects in cast aluminium using X-radioscopic images. The first method is based on the assumption that most defects have the shape of a circular high-intensity spot. Therefore, defects are detected using a template matching-like algorithm. This method works well when the defects are far enough from the edges of the major shapes in the image, and when the image gives a closer view of the defect. The second method deals with the defects which are closer to the edges in the image, and therefore are missed by the first method. This method distinguishes between defects and edges by using the following properties of a defect: they are local maxima of the image intensity, and the distribution of the intensity in a patch around the defect should resemble more that of a corner than that of an edge. Both local maxima and corner-like properties are computed using the second order derivatives of the image intensities, and the Harris Corner Detector algorithm. The third algorithm is a simple combination of the aforementioned methods in which a pixel is considered to be a defect if it is detected as a defect by either of the two methods. We present experiments using the third method showing that 94.3% of the defects are correctly detected, with only 1.3 false alarms per image.
  • Thumbnail Image
    Item