Adaptively Denoising Proposal Collection for Weakly Supervised Object Localization

Abstract

In this paper, we address the problem of weakly supervised object localization (WSL), which trains a detection network on the dataset with only image-level annotations. The proposed approach is built on the observation that the proposal set from the training dataset is a collection of background, object parts, and objects. Several strategies are taken to adaptively eliminate the noisy proposals and generate pseudo object-level annotations for the weakly labeled dataset. A multiple instance learning (MIL) algorithm enhanced by mask-out strategy is adopted to collect the class-specific object proposals, which are then utilized to adapt a pretrained classification network to a detection network. In addition, the detection results from the detection network are re-weighted by jointly considering the detection scores and the overlap ratio of proposals in a proposal subset optimization framework. The optimal proposals work as object-level labels that enable a pseudo-strongly supervised dataset for training the detection network. Consequently, we establish a fully adaptive detection network. Extensive evaluations on the PASCAL VOC 2007 and 2012 datasets demonstrate a significant improvement compared with the state-of-the-art methods.

Colonoscopy polyp detection and classification: Dataset creation and comparative evaluations

“…Deep learning models can automatically generate parameters with deeper layers and extract high-level semantic features. Especially in recent years, many new models [ 37 , 38 ] and techniques [ 39 – 42 ] have been published to set new records in various computer vision tasks. [ 43 ] employs multi-scale architecture with 3 layers of CNN and 3 layers of max-pooling followed by fully connected layers.…”

A Domain Gap Aware Generative Adversarial Network for Multi-domain Image Translation

“…Deep neural networks have shown great potential in dealing with real-world tasks [11], [12], [13], [14], [15], [16], [17]. Many deep learning based methods were proposed for image content understanding [18], [19] and image content generation tasks [20], [21], [22].…”

Enhanced U-Net: A Feature Enhancement Network for Polyp Segmentation

“…However, these approaches are slow and have a high misdetection rate due to the low representation capability of hand-crafted features. Following the huge success of deep learning-based models on generic datasets [7] [8] [9] [10], researchers started using neural networks for polyp detection and segmentation. Inspired by the early work [11], where FCN [13] is utilized with a pre-trained model to segment the polyp, Akbari et al [12] proposed a modified version of FCN to improve the performance of polyp segmentation.…”