Cross-Domain Weakly-Supervised Object Detection Through Progressive Domain Adaptation

By Naoto Inoue, Ryosuke Furuta, Toshihiko Yamasaki, Kiyoharu Aizawa. 2018

629 citations recorded by the source corpus.

Can we detect common objects in a variety of image domains without instance-level annotations? In this paper, we present a framework for a novel task, cross-domain weakly supervised object detection, which addresses this question. For this paper, we have access to images with instance-level annotations in a source domain (e.g., natural image) and images with image-level annotations in a target domain (e.g., watercolor). In addition, the classes to be detected in the target domain are all or a subset of those in the source domain. Starting from a fully supervised object detector, which is pre-trained on the source domain, we propose a two-step progressive domain adaptation technique by fine-tuning the detector on two types of artificially and automatically generated samples. We test our methods on our newly collected datasets1 containing three image domains, and achieve an improvement of approximately 5 to 20 percentage points in terms of mean average precision (mAP) compared to the best-performing baselines.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Prototype-Guided Continual Adaptation for Class-Incremental Unsupervised Domain Adaptation (2022, background): “Step 2 [10,11,12,13,14,15,16,17,18,19] Kettle, Folder, Keyboard, Flipflops, Pencil, Bed, Hammer, ToothBrush, Couch, Bike”
Unlabeled Data Improves Adversarial Robustness (2019, neutral): “[19] N.”
Unpaired Cross-Spectral Pedestrian Detection Via Adversarial Feature Learning (2019, methodology): “They are also applied to several practical applications such as object detection [15] and semantic segmentation [16].”
Unbiased Mean Teacher for Cross-domain Object Detection (2020, background): “(a) and (c) are respectively a source example image from the PASCAL VOC dataset [8], and a target example image from the Clipart1k dataset [19].”
Few-Shot Adaptive Faster R-CNN (2019, methodology): “The method has been used as a module in [20] for adapting object detector to target domain.”
Confidence Regularized Self-Training (2019, background): “More recently, self-training with networks emerged as a promising alternative towards domain adaptation [4, 5, 25, 29, 49, 54, 69].”
Cross-Domain Adaptive Teacher for Object Detection (2021, methodology): “We conduct our experiments on five public datasets, including Cityscapes [4]), Foggy Cityscapes [33], PASCAL VOC [9], Clipart1k [16], and Watercolor2k [16].”
Minimum Class Confusion for Versatile Domain Adaptation (2019, methodology): “Some methods [39,16,55,56] also utilize the less-reliable self-training or pseudo labeling, e.”
Self-Training: A Survey (2022, background): “For instance, in the domain of computer vision, a pre-text task might involve rotating images within a predeﬁned range of angles, followed by training a supervised model to predict these angles.”
Multi-task incremental learning for object detection (2020, methodology): “We consider different combinations of the following datasets: Watercolor [18], Clipart [18], Comic [18], Kitchen [11], KITTI [10], VOC [9], COCO [23].”
Deep Visual Domain Adaptation: A Survey (2018, methodology): “By using bounding box labeled data in a source domain and weak labeled data in a target domain, [55] progressively ﬁne-tuned the pre-trained model with domain-transfer samples and pseudo-labeling samples.”
Adapting Object Detectors via Selective Cross-Domain Alignment (2019, methodology): “Note that other related works, are either with weakly supervised settings [21] or with class-speciﬁc settings [35], and cannot be directly compared.”
Object Detection in 20 Years: A Survey (2019, neutral): “Cycle-consistent transformation [222] has also been applied to bridge the gap between source and target domains [223], [224].”
Towards Universal Object Detection by Domain Attention (2019, methodology): “The 20 VOC categories are replicated on CrossDomain [17] with three subsets of Watercolor, Clipart and Comic, with objects depicted in watercolor, clipart and comic styles, respectively.”
Harmonizing Transferability and Discriminability for Adapting Object Detectors (2020, methodology): “Clipart is from the Watercolor datasets [21] and used as the target domain, which contains 1K images and have the same 20 categories as PASCAL.”
Weakly Supervised Object Localization and Detection: A Survey (2021, neutral): “The examples are from [24], [53], [69], where detection results in different colors in the painting images indicate different types of objects. the normal image and abnormal image is the uniﬁed distribution, the image data can then be decomposed by the low-rank subspace learning technique to obtain…”
Category Anchor-Guided Unsupervised Domain Adaptation for Semantic Segmentation (2019, methodology): “To overcome this issue, a variety of domain adaptation methods for classiﬁcation [6, 23, 37, 28, 41, 3, 19], detection [38, 16], and segmentation [7, 14, 13, 27, 34, 40, 21, 47] have been proposed.”