2018/03/30 by Naoto Inoue, Inoue, Naoto, Ryosuke Furuta +5 · 9 citations
Computer Science · #Advanced Neural Network Applications #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Multimodal Machine Learning Applications
paper · pdf · doi:10.48550/arxiv.1803.11365
openalex publication_date 2018/03/30 · openalex created_date 2021/10/11 · openalex updated_date 2026/07/28
Can we detect common objects in a variety of image domains without\ninstance-level annotations? In this paper, we present a framework for a novel\ntask, cross-domain weakly supervised object detection, which addresses this\nquestion. For this paper, we have access to images with instance-level\nannotations in a source domain (e.g., natural image) and images with\nimage-level annotations in a target domain (e.g., watercolor). In addition, the\nclasses to be detected in the target domain are all or a subset of those in the\nsource domain. Starting from a fully supervised object detector, which is\npre-trained on the source domain, we propose a two-step progressive domain\nadaptation technique by fine-tuning the detector on two types of artificially\nand automatically generated samples. We test our methods on our newly collected\ndatasets containing three image domains, and achieve an improvement of\napproximately 5 to 20 percentage points in terms of mean average precision\n(mAP) compared to the best-performing baselines.\n