Matrix Completion for Weakly-Supervised Multi-Label Image Classification

Abstract

Abstract-In the last few years, image classification has become an incredibly active research topic, with widespread applications. Most methods for visual recognition are fully supervised, as they make use of bounding boxes or pixelwise segmentations to locate objects of interest. However, this type of manual labeling is time consuming, error prone and it has been shown that manual segmentations are not necessarily the optimal spatial enclosure for object classifiers. This paper proposes a weakly-supervised system for multi-label image classification. In this setting, training images are annotated with a set of keywords describing their contents, but the visual concepts are not explicitly segmented in the images.We formulate the weakly-supervised image classification as a low-rank matrix completion problem. Compared to previous work, our proposed framework has three advantages: (1) Unlike existing solutions based on multiple-instance learning methods, our model is convex. We propose two alternative algorithms for matrix completion specifically tailored to visual data, and prove their convergence. (2) Unlike existing discriminative methods, our algorithm is robust to labeling errors, background noise and partial occlusions. (3) Our method can potentially be used for semantic segmentation. Experimental validation on several datasets shows that our method outperforms state-of-the-art classification algorithms, while effectively capturing each class appearance.

Adapting Convolutional Neural Network to Multi-label Image Classification

“…We compared AUROC results among MC-Pos, MC-Simplex [2] with three approaches: Endto-End Training (EET), Pre-training and Fine-tuning (PF) and AdaBoost.MH with CNN features (ACNN). Three approaches ran on 'GeForce GTX TITAN X' GPU platform with Torch7 toolkit except that AdaBoost.MH was implemented by Java.…”

“…In the multi-label image classification, a image can be categorized as multiple semantic categories [1]. Cabral et al [2] solved the multi-label image classification problems with matrix completion. Li et al [7] used informative label combination pairs to augment the original labels to enhance the individual label prediction.…”

Noise-Clustered Distant Supervision for Relation Extraction: A Nonparametric Bayesian Perspective

“…The essence of the task is a multi-label classification problem (Cabral et al, 2011) with noisy patterns (Han and Sun, 2014). One simple way, to solve the problem, is to learn separate classifiers for each of relation labels, using n samples with d features, by optimizing b ∈ R 1×1 and w ∈ R d×1 ,…”

“…However, label correlations are not considered in the above formulation. To jointly consider feature correlations and label correlations, (Cabral et al, 2011) formulated the multi-label classification as a matrix completion problem. As a powerful framework, it has been successfully applied to relation extraction task with distant supervision.…”

“…The work in (Fan et al, 2014) first adopted the mentioned framework, as a general joint learning and inference framework (Cabral et al, 2011), to learn noise-tolerant distant supervision for relation extraction. It achieves the state-of-the-art performance.…”