Supplementary Material - Optimizing Average Precision using Weakly Supervised Data

نویسندگان

  • Aseem Behl
  • C. V. Jawahar
  • M. Pawan Kumar
چکیده

Briefly, the algorithm starts by specifying no constraints (step 1 of Algorithm 1: W is initialized to the null set). At each iteration, it adds a single constraint, which corresponds to the most violated ranking (step 4 of Algorithm 1: solving problem (1)). Intuitively, problem (1) finds a ranking that differs significantly from the ground-truth ranking in terms of AP but has a high score. Having added the most violated constraint, the cutting plane algorithm updates the parameters by solving a convex quadratic program (step 3 of Algorithm 1). The algorithm stops once no constraint can be found that is violated by more than the desired precision . The feasibility of the cutting plane algorithm relies on solving problem (1) efficiently. For fixed values of HP and HN (which is indeed the case for supervised AP-SVM), this can be achieved using the algorithm described in the following appendix.

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Weakly Supervised Object Boundaries Supplementary material

In this work we propose to train boundary detectors using weakly supervised annotations. We propose and evaluate multiple strategies to generate annotations fusing different sources, such as unsupervised image segmentation [2], object proposal methods [10, 5], and object detectors [3, 6] (trained on bounding boxes). Figure 5 illustrates the examples of the proposed weakly supervised boundary an...

متن کامل

Weakly Supervised Object Localization with Progressive Domain Adaptation Supplementary Material

In this supplementary material, we present three additional results to complement the paper. First, we report detailed quantitative evaluation on the PASCAL VOC and ILSVRC object detection datasets. Second, we show additional qualitative detection results on the VOC 2007 dataset. Third, we analyze the errors of three variants of the proposed approach and show relative contributions from each co...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

عنوان ژورنال:

دوره   شماره 

صفحات  -

تاریخ انتشار 2014