Multinomial latent logistic regression
Zhe Xu · UTS ePRESS (University of Technology Sydney) · 2016
We are arriving at the era of big data.The booming of data gives birth to more complicated research objectives, for which it is important to utilize the superior discriminative power brought by explicitly designed feature representations.However, training models based on these features usually requires detailed human annotations, which is being intractable due to the exponential growth of data scale.A possible solution for this problem is to employ a restricted form of training data, while regarding the others as latent variables and performing latent variable inference during the training process.This 5.2 Localization results for CUB-200-2010, CUB-200-2011 and Stanford Dogs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .93 5.3 A detailed comparison with baselines of different localization strategies and classification methods on the CUB-200-2010 dataset.Row 1-3 show results by training classifiers solely on the detected foreground regions.Row 4-6 show results by performing a multiinstance learning (MIL) approach initialized by the respective localization results.The final row presents an upper bound of our algorithm by using ground-truth bounding box supervision. . . . .95 5.4 Effect of fine-tuning CNNs.We achieved an accuracy of 77.37% on the CUB-200-2011 dataset under the weakly supervised scenario.96 5.5 Performance comparison to the state-of-the-art results in the literature with or without the use of ground-truth bounding boxes at the training stage. . . . . . . . . . . . . . . . . . . . . . . . . . . .