Learning Predictable and Discriminative Attributes for Visual Recognition

Authors: Yuchen Guo, Guiguang Ding, Xiaoming Jin, Jianmin Wang

AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive experiments on Animals with Attributes (AwA) and Caltech256 datasets, and the results demonstrate that the proposed method achieves state-of-the-art performance.
Researcher Affiliation Academia Yuchen Guo and Guiguang Ding and Xiaoming Jin and Jianmin Wang School of Software, Tsinghua University, Beijing 100084, China yuchen.w.guo@gmail.com, {dinggg,xmjin,jimwang}@tsinghua.edu.cn,
Pseudocode Yes Algorithm 1 Learning Algorithm Input: Image data X, labels Y, #attributes k. Output: Two groups of classifiers, wva and wac. 1: Initialization: A PCA(X, k), 2: Binarization: A sign(A). 3: Construct W by Eq. (6), a diagonal matrix with diagonal element Dii = Pn j=1 Wij and L = D W. 4: Training classifiers: learn SVM classifiers wva and wac. 5: repeat 6: Optimize A greedily to minimize Eq. (8) by block coordinate descent algorithm. 7: Update classifiers wva and wac. 8: until Convergence 9: Return wva and wac.
Open Source Code No The paper does not provide any concrete access information (e.g., repository link, explicit statement of code release) for the source code of the described methodology.
Open Datasets Yes The Animal with Attributes (AwA) dataset (Lampert, Nickisch, and Harmeling 2009) contains 30, 475 images from 50 animal categories. ... In this paper, we conduct image retrieval experiments with binary attributes (or hash codes) on Caltech256 dataset.
Dataset Splits Yes Following the setting in previous works, we change the numbers of images from each category (15, ..., 50) when training models, and select other 10 images per category for validation. Then all of the rest images form the test set.
Hardware Specification Yes All results above are obtained on a computer which has Intel Core i7-2600 3.40GHz CPU and 8GB RAM.
Software Dependencies No The paper mentions various techniques and classifiers (e.g., SVM classifiers, PCA, SIFT, LLC) but does not provide specific version numbers for any software, libraries, or frameworks used in the experiments.
Experiment Setup Yes And for PDA, we consistently set λ = 1 to balance predictability and discriminability, α = 0.1 for intra-category locality regularization, p to 4 times of the number of training images per category for constructing nearest neighbor matrix W, and the classifier regularization parameters are all set to 0.1.