Self Paced Deep Learning for Weakly Supervised Object Detection

Abstract

In a weakly-supervised scenario object detectors need to be trained using image-level annotation alone. Since bounding-box-level ground truth is not available, most of the solutions proposed so far are based on an iterative, Multiple Instance Learning framework in which the current classifier is used to select the highest-confidence boxes in each image, which are treated as pseudo-ground truth in the next training iteration. However, the errors of an immature classifier can make the process drift, usually introducing many of false positives in the training dataset. To alleviate this problem, we propose in this paper a training protocol based on the self-paced learning paradigm. The main idea is to iteratively select a subset of images and boxes that are the most reliable, and use them for training. While in the past few years similar strategies have been adopted for SVMs and other classifiers, we are the first showing that a self-paced approach can be used with deep-network-based classifiers in an end-to-end training pipeline. The method we propose is built on the fully-supervised Fast-RCNN architecture and can be applied to similar architectures which represent the input image as a bag of boxes. We show state-of-the-art results on Pascal VOC 2007, Pascal VOC 2010 and ILSVRC 2013. On ILSVRC 2013 our results based on a low-capacity AlexNet network outperform even those weakly-supervised approaches which are based on much higher-capacity networks.

To appear at IEEE Transactions on PAMI

References in corpus (8)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• FaceNet: A Unified Embedding for Face Recognition and Clustering
• Return of the Devil in the Details: Delving Deep into Convolutional Nets
• STC: A Simple to Complex Framework for Weakly-supervised Semantic Segmentation
• Weakly Supervised Object Localization with Multi-fold Multiple Instance Learning
• Weakly-supervised Discovery of Visual Pattern Configurations
• Bridging Saliency Detection to Weakly Supervised Object Detection Based on Self-paced Curriculum Learning
• Computational Baby Learning

Cited by in corpus (16)

• MentorNet: Learning Data-Driven Curriculum for Very Deep Neural Networks on Corrupted Labels
• Credit spread approximation and improvement using random forest regression
• Self-learning for weakly supervised Gleason grading of local patterns
• Self-Training and Adversarial Background Regularization for Unsupervised Domain Adaptive One-Stage Object Detection
• Learning Self-Awareness for Autonomous Vehicles: Exploring Multisensory Incremental Models
• Deep traffic light detection by overlaying synthetic context on arbitrary natural images
• A Self-paced Regularization Framework for Partial-Label Learning
• Weakly Supervised 3D Object Detection from Point Clouds
• Towards Human-Machine Cooperation: Self-supervised Sample Mining for Object Detection
• Self-Paced Multi-Label Learning with Diversity
• Self-Paced Probabilistic Principal Component Analysis for Data with Outliers
• Low-Shot Learning from Imaginary 3D Model
• Curriculum Learning with Diversity for Supervised Computer Vision Tasks
• Progressive Stage-wise Learning for Unsupervised Feature Representation Enhancement
• Towards Accurate Localization by Instance Search
• Weakly Supervised Object Localization and Detection: A Survey