LSDA: Large Scale Detection through Adaptation

By Judy Hoffman, Sergio Guadarrama, Eric Tzeng, Ronghang Hu, Jeff Donahue, Ross B. Girshick, Trevor Darrell, Kate Saenko. 2014

338 citations recorded by the source corpus.

A major challenge in scaling object detection is the difficulty of obtaining labeled images for large numbers of categories. Recently, deep convolutional neural networks (CNNs) have emerged as clear winners on object classification benchmarks, in part due to training with 1.2M+ labeled classification images. Unfortunately, only a small fraction of those labels are available for the detection task. It is much cheaper and easier to collect large quantities of image-level labels from search engines than it is to collect detection data and label it with precise bounding boxes. In this paper, we propose Large Scale Detection through Adaptation (LSDA), an algorithm which learns the difference between the two tasks and transfers this knowledge to classifiers for categories without bounding box annotated data, turning them into detectors. Our method has the potential to enable detection for the tens of thousands of categories that lack bounding box annotations, yet have plenty of classification data. Evaluation on the ImageNet LSVRC-2013 detection challenge demonstrates the efficacy of our approach. This algorithm enables us to produce a >7.6K detector by using available classification data from leaf nodes in the ImageNet tree. We additionally demonstrate how to modify our architecture to produce a fast detector (running at 2fps for the 7.6K detector). Models and software are available at lsda.berkeleyvision.org.

Subfields: 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 Computer Vision and Pattern Recognition

Citations in the corpus

Domain randomization for transferring deep neural networks from simulation to the real world (2017, background): “The computer vision community has devoted signiﬁcant study to the problem of adapting vision-based models trained in a source domain to a previously unseen target domain (see, e.g., [9], [14], [15], [19], [23], [25], [51]).”
Learning Open-World Object Proposals without Learning to Classify (2021, background): “Approaches under semisupervised setup [29], [30], [31], [32] employ the weak imagelevel labels for novel classes as well as box-level labels for base classes.”
A dataset for Movie Description (2015, methodology): “Additionally we present an approach to semi-automatically collect and align DVS data and analyse the differences between DVS and movie scripts.”
Natural Language Object Retrieval (2015, methodology): “Significant progress has been made in object detection in recent years; with the help of Convolutional Neural Networks (CNNs), it is possible to detect a predefined set of object categories with high accuracy [8, 7], and the number of categories in object detection has grown over 10K to 100K with th”
Learning Transferable Features with Deep Adaptation Networks (2015, methodology): “, 2012), a strong model when it is adapted to novel tasks (Donahue et al., 2014; Hoffman et al., 2014).”
Unsupervised Domain Adaptation with Residual Transfer Networks (2016, background): “…goal of this paper is to design a deep neural network that enables learning of transfer classifiers y = fs (x) and y = ft (x) to close the source-target discrepancy, such that the expected target risk Rt (ft) = E(x,y)∼q [ft (x) 6= y] can be bounded by leveraging the source domain supervised data.”
Cross Modal Distillation for Supervision Transfer (2015, methodology): “with the introduction of supervised CNN models by Krizhevsky et al. [31], the community has been moving towards a generic set of features which can be specialized to speciﬁc tasks and domains at hand [7,13,12, 40,23] and traditional visual adaptation techniques can be used in conjunction with such f”
(CAD)$^2$RL: Real Single-Image Flight without a Single Real Image (2016, background): “domain adaptation methods have been proposed that aim to generalize perception systems trained in a source domain into a target domain [42, 17].”
Simultaneous Deep Transfer Across Domains and Tasks (2015, background): “With limited labels on a subset of the categories, the traditional domain transfer approach of ﬁne-tuning on the available target data [14, 29, 17] is not effective.”
Deep Domain Confusion: Maximizing for Domain Invariance (2014, background): “It has been shown that a pre-trained CNN can be adapted for a new task through ﬁne-tuning [15, 30, 18].”
Weakly-and Semi-Supervised Learning of a Deep Convolutional Network for Semantic Image Segmentation (2015, background): “Combining few strong annotations and a large number of weak annotations in a semisupervised setting is another emerging research direction (Hoffman et al., 2014).”
Doctor AI: Predicting Clinical Events via Recurrent Neural Networks (2015, methodology): “To address this challenge, we resort to the recent advances in domain adaptation techniques for deep neural networks [33, 3, 50, 18, 37].”
Deep Compositional Captioning: Describing Novel Object Categories without Paired Training Data (2015, background): “In [12], zero-shot object detectors are learned by transferring information about how network weights trained on the classification task differ from weights trained on the detection task.”
Towards a Visual Turing Challenge (2014, background): “Architectures that aim at mimicking or reproducing this space of human concepts need to capture the same diversity and therefore scale up to thousands of concepts [38, 39, 40].”
Open-Vocabulary Object Detection Using Captions (2020, methodology): “Particularly, we compare to LSDA [18], which learns a transformation from classifier weights into detector weights, its extension [36] to utilize semantic class relationships (LSDA+), and a more recent work [37] which uses multiple instance learning along with a region proposal network trained on ba”
Domain Adaptation for Visual Applications: A Comprehensive Survey (2017, methodology): “Indeed, the pre-trained deep model combined with class-agnostic region of interest proposals, can be used to predict the presence or absence of the object of interest in the proposed local regions of interest [190, 99, 157, 114].”
Multi-Adversarial Domain Adaptation (2018, background): “Thus deep networks have been explored for transfer learning (Glorot, Bordes, and Bengio 2011; Oquab et al. 2013; Hoffman et al. 2014), multimodal and multi-task learning (Collobert et al.”
FCNs in the Wild: Pixel-level Adversarial and Constraint-based Adaptation (2016, methodology): “In detection, Hoffman et al . proposed a domain adaptation system by explicitly modeling the representation shift be-tween classiﬁcation and detection models [11] along with a follow-up work which incorporated per-category adaptation using multiple instance learning [12].”