Semi-Supervised Self-Training of Object Detection Models

By Chuck Rosenberg, Martial Hebert, Henry Schneiderman. 2005

897 citations recorded by the source corpus.

The construction of appearance-based object detection systems is time-consuming and difficult because a large number of training examples must be collected and manually labeled in order to capture variations in object appearance. Semi-supervised training is a means for reducing the effort needed to prepare the training set by training the model with a small number of fully labeled examples and an additional set of unlabeled or weakly labeled examples. In this work we present a semi-supervised approach to training object detection systems based on self-training. We implement our approach as a wrapper around the training process of an existing object detector and present empirical results. The key contributions of this empirical study is to demonstrate that a model trained in this manner can achieve results comparable to a model trained in the traditional manner using a much larger set of fully labeled data, and that a training data selection metric that is defined independently of the detector greatly outperforms a selection metric based on the detection confidence generated by the detector.

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

FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence (2020, methodology): “, [44]).”
Towards Interpretable Natural Language Understanding with Explanations as Latent Variables (2020, methodology): “…(ours) 42.5 66.4 In the semi-supervised setting , we compare ELV-EST against several competitive semi-supervised text classiﬁcation methods including Pseudo-Labeling [29], Self-Training [30], and Data Programming [13] which incorporates NL explanations to perform semi-supervised text classiﬁcation.”
Semi-supervised Learning with Deep Generative Models (2014, methodology): “Amongst existing approaches, the simplest algorithm for semi-supervised learning is based on a self-training scheme (Rosenberg et al., 2005) where the the model is bootstrapped with additional labelled data obtained from its own highly confident predictions; this process being repeated until some te”
Self-Training for End-to-End Speech Recognition (2019, methodology): “Self-training has been applied to tasks in natural language processing including word-sense disambiguation [2], noun identification [23] and parsing [3], in addition to tasks in computer vision such as object detection [24, 25] and image classification [26].”
Meta-Learning for Semi-Supervised Few-Shot Classification (2018, background): “As for the literature on semi-supervised learning, while it is quite vast (Zhu, 2005; Chapelle et al., 2010), the most relevant category to our work is related to self-training (Yarowsky, 1995; Rosenberg et al., 2005).”
Rethinking Pre-training and Self-training (2020, background): “Closer in applications are semi-supervised learning for detection and segmentation (e.g., [44–47]), who only study self-training in isolation or without a comparison against ImageNet pre-training.”
Data Distillation: Towards Omni-Supervised Learning (2017, methodology): “Among semi-supervised methods, our method is most related to self-training, a strategy in which a model’s predictions on unlabeled data are used to train itself [36, 48, 43, 33, 22, 46, 5, 21].”
Unlabeled Data Improves Adversarial Robustness (2019, methodology): “We can use such classifiers on the unlabeled data and obtain useful pseudo-labels, which directly suggests the use of self-training—one of the oldest frameworks for semi-supervised learning [32], which consists of applying a supervised training method on the pseudo-labeled data.”
ReMixMatch: Semi-Supervised Learning with Distribution Alignment and Augmentation Anchoring (2019, methodology): “Related to this idea are “self-training” methods (McLachlan, 1975; Rosenberg et al., 2005) such as Pseudo-Label (Lee, 2013) that use the predicted class on an unlabeled input as a hard target for the same input, which implicitly minimizes the entropy of the prediction.”
Domain Adaptation for Visual Applications: A Comprehensive Survey (2017, methodology): “Using these new samples the model can be updated by using semi-supervised techniques such as self-training [180, 248], or co-training [119, 133].”
Watch and learn: Semi-supervised learning of object detectors from videos (2015, methodology): “Decorrelated errors: To remove high conﬁdence false positives (see Figure 3), we rely on the principle of decorre-lated errors (similar to multi-view SSL [43, 47]).”
Audio-Visual Instance Discrimination with Cross-Modal Agreement (2020, background): “CMA builds upon insights from prior work [70] in multi-view learning.”
Neural Snowball for Few-Shot Relation Learning (2019, methodology): “Many other semi-supervised methods can also be adopted for RE (Rosenberg, Hebert, and Schneiderman 2005; French, Mackiewicz, and Fisher 2017; Lin et al. 2019), yet they still require sufficient annotations and mainly aim at classifying predefined relations rather than discovering new ones.”
Smooth Neighbors on Teacher Graphs for Semi-Supervised Learning (2017, methodology): “Self-training methods iteratively use the current classifier to label those unlabeled ones with high confidence [29].”
Semi-supervised tuning from temporal coherence (2015, background): “In particular, in self-training (Rosenberg et al. (2005)), a classiﬁer is ﬁrst trained with a small amount of labeled data and then used to classify the unlabeled data.”
IncDet: In Defense of Elastic Weight Consolidation for Incremental Object Detection (2020, methodology): “Pseudo annotation proposed in this paper is reminiscent of self-training which is widely used in general machine learning, as well as object detection [54].”
Soft Sampling for Robust Object Detection (2018, positive): “ey would not increase our pool of positive labels. Harvesting more training data using a trained classiﬁer is known to provide a bump in performance. This is commonly done in semi-supervised learning [16]. Unfortunately, in the large scale setting, current detectors do not perform very well and thei”