A Simple Semi-Supervised Learning Framework for Object Detection

By Kihyuk Sohn, Zizhao Zhang, Chun-Liang Li, Chen-Yu Lee, Tomas Pfister. 2020

598 citations recorded by the source corpus.

Semi-supervised learning (SSL) has promising potential for improving the predictive performance of machine learning models using unlabeled data. There has been remarkable progress, but the scope of demonstration in SSL has been limited to image classification tasks. In this paper, we propose STAC, a simple yet effective SSL framework for visual object detection along with a data augmentation strategy. STAC deploys highly confident pseudo labels of localized objects from an unlabeled image and updates the model by enforcing consistency via strong augmentations. We propose new experimental protocols to evaluate performance of semi-supervised object detection using MS-COCO and demonstrate the efficacy of STAC on both MS-COCO and VOC07. On VOC07, STAC improves the AP$^{0.5}$ from 76.30 to 79.08; on MS-COCO, STAC demonstrates 2x higher data efficiency by achieving 24.38 mAP using only 5% labeled data than supervised baseline that marks 23.86% using 10% labeled data. The code is available at \url{this https URL}.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Representation & Self-Supervised Learning

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

In Representation & Self-Supervised Learning

Citations in the corpus

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.”
End-to-End Semi-Supervised Object Detection with Soft Teacher (2021, methodology): “0 mAP on val2017 with 1%, 5% and 10% labeled data using the Faster R-CNN [22] framework with ResNet-50 [8] and FPN [14], surpassing previous best method STAC [27] by +6.”
SODA10M: Towards Large-Scale Object Detection Benchmark for Autonomous Driving (2021, methodology): “com/danielgordon10/vince STAC [47] https://github.”
A Survey on Deep Semi-Supervised Learning (2021, background): “Applications Datasets Citations Image classification MNIST [51], SVHN [52], STL-10 [53], Cifar-10,Cifar-100 [54], ImageNet [55] [56], [57], [58], [59], [60], [61], [62], [63], [64] Object detection PASCAL VOC [65], MSCOCO [66], ILSVRC [67] [49], [68], [69], [70] 3D object detection SUN RGB-D [71], S”
PreDet: Large-scale weakly supervised pre-training for detection (2021, methodology): “Semi-supervised learning and Self-training Semisupervised and self-training methods [50, 62, 22, 39, 29, 70, 60, 61] train a model jointly with a specific target dataset and the provided large-scale dataset, and have shown benefits for detection.”
PseudoSeg: Designing Pseudo Labels for Semantic Segmentation (2020, methodology): “Successful examples include MeanTeacher (Tarvainen & Valpola, 2017), UDA (Xie et al., 2019), MixMatch (Berthelot et al., 2019b), FeatMatch (Kuo et al., 2020), and FixMatch (Sohn et al., 2020a).”
ST++: Make Self-trainingWork Better for Semi-supervised Semantic Segmentation (2021, methodology): “Among them, the most related ones to us are [48, 55].”
Unsupervised Prompt Learning for Vision-Language Models (2022, unknown): “To answer this question, we perform the following calculations on the UCF101 dataset: 1) per-class pseudo label accuracy; 2) per-class transfer improvement (accuracy difference between UPL and original CLIP).”
Propagate Yourself: Exploring Pixel-Level Consistency for Unsupervised Visual Representation Learning (2020, methodology): “We evaluate feature transfer performance on four downstream tasks: object detection on Pascal VOC [15], object detection on COCO [23], semantic segmentation on Cityscapes [10], and semi-supervised object detection on COCO [28].”
Semi-Supervised and Unsupervised Deep Visual Learning: A Survey (2022, methodology): “A simple self-training paradigm is to annotate the unlabeled images with bounding boxes and their class labels using a pre-trained teacher model and use these data for re-training [243].”
DetCLIPv2: Scalable Open-Vocabulary Object Detection Pre-training via Word-Region Alignment (2023, methodology): “To utilize the image-text pair data without instance-level annotation, approaches [13, 14, 19, 27, 48, 54] generate pseudo-boundingbox labels following a self-training paradigm [40] or based on a pre-trained VL model [33].”
Scaling Open-Vocabulary Object Detection (2023, methodology): “In self-training, an existing detector is used to predict bounding boxes on unlabeled images to generate data for training better detectors [25, 42, 29].”
Detecting Twenty-thousand Classes using Image-level Supervision (2022, methodology): “We compare with the following ways to use image-level supervision: Self-training applies the Box-Supervised model to the image-classification data and collects all predictions with score > 0.5 as pseudo-labels for finetuning.”