Few-Shot Object Detection via Classification Refinement and Distractor Retreatment

Li, Yiting; Zhu, Haiyue; Cheng, Yu; Wang, Wenxin; Teo, Chek Sing; Chen, Xiang; Vadakkepat, Prahlad; Lee, Tong Heng

doi:10.1109/cvpr46437.2021.01514

Cited by 72 publications

(34 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…And due to other factors like the chosen of detection framework, the performance of this method is not quite satisfactory. CRaDR [45] proposes a Few-Shot Correction Network (FRCN), which extracts features of region proposals individually and generates classification scores for them. These classification scores are used to refine the classification scores of the main model.…”

Section: Transfer-learning Methodsmentioning

confidence: 99%

A Survey of Deep Learning for Low-Shot Object Detection

Huang¹,

Zhang²,

Xue³

et al. 2021

Preprint

View full text Add to dashboard Cite

Object detection is a fundamental task in computer vision and image processing. Current deep learning based object detectors have been highly successful with abundant labeled data. But in real life, it is not guaranteed that each object category has enough labeled samples for training. These large object detectors are easy to overfit when the training data is limited. Therefore, it is necessary to introduce few-shot learning and zeroshot learning into object detection, which can be named low-shot object detection together. Low-Shot Object Detection (LSOD) aims to detect objects from a few or even zero labeled data, which can be categorized into few-shot object detection (FSOD) and zero-shot object detection (ZSD), respectively. This paper conducts a comprehensive survey for deep learning based FSOD and ZSD. First, this survey classifies methods for FSOD and ZSD into different categories and discusses the pros and cons of them. Second, this survey reviews dataset settings and evaluation metrics for FSOD and ZSD, then analyzes the performance of different methods on these benchmarks. Finally, this survey discusses future challenges and promising directions for FSOD and ZSD.

show abstract

Section: Transfer-learning Methodsmentioning

confidence: 99%

A Survey of Deep Learning for Low-Shot Object Detection

Huang¹,

Zhang²,

Xue³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Recent research based on meta-learning [39] has also obtained remarkable performance [4], [5], [6], [7], [9], resulting in better generalization and faster deployment than transfer learning. Low-shot classification correction network (LSCN) [8] proposes classification refinement with four different parts (unified recognition, global receptive field, interclass separation, and confidence calibration) to boost the performance of the overall classes. Apart from the use of meta-learning, openended centre net (ONCE) [9] built on the CenterNet [27] first proposes a new study of incremental few-shot object detection setting, where the new classes are registered incrementally without using the samples from base classes.…”

Section: Related Workmentioning

confidence: 99%

“…Few-shot object detection is a trending research topic aiming at training object detectors that generalize well with a small amount of object annotations. Studies have shown that directly applying DNNs designed for big datasets to few-shot object detection tasks often leads to overfitting [1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12]. Various learning strategies, such as meta-learning [4], [5], [6], [7] and transfer learning [1], [2], [3], have been explored to address this issue.…”

Section: Introductionmentioning

confidence: 99%

“…Various learning strategies, such as meta-learning [4], [5], [6], [7] and transfer learning [1], [2], [3], have been explored to address this issue. Specifically, by combining some state-of-the-art methods gaining significant improvement in object detection, such studies [8], [9], [10] succeed to boosting the performance of few-shot object detection. More recently, any-shot object detector [11] proposes a realistic setting for the unseen and few-shot novel categories and weakly any-shot object detector [12] extend the former any-shot object detector with weakly supervision.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Leveraging Bottom-Up and Top-Down Attention for Few-Shot Object Detection

Chen¹,

Jiang²,

Zhao³

2020

Preprint

View full text Add to dashboard Cite

Few-shot object detection aims at detecting objects with few annotated examples, which remains a challenging research problem yet to be explored. Recent studies have shown the effectiveness of self-learned top-down attention mechanisms in object detection and other vision tasks. The top-down attention, however, is less effective at improving the performance of few-shot detectors. Due to the insufficient training data, object detectors cannot effectively generate attention maps for few-shot examples. To improve the performance and interpretability of few-shot object detectors, we propose an attentive few-shot object detection network (AttFDNet) that takes the advantages of both top-down and bottom-up attention. Being task-agnostic, the bottom-up attention serves as a prior that helps detect and localize naturally salient objects. We further address specific challenges in few-shot object detection by introducing two novel loss terms and a hybrid few-shot learning strategy. Experimental results and visualization demonstrate the complementary nature of the two types of attention and their roles in few-shot object detection. Codes are available at https://github.com/chenxy99/AttFDNet.

show abstract

“…In addition, intelligent systems like human brains do not need millions of samples to learn useful patterns and are energy-efficient. On the premise of it, learning with small data has been an important research area in various tasks [7,16,19,28,31,35,37,41,44]. Among numerous promising works along the direction, a limited amount target on generative models.…”

Section: Introductionmentioning

confidence: 99%

Implicit Data Augmentation Using Feature Interpolation for Low-Shot Image Generation

Dai¹,

Hang²,

Guo³

2021

Preprint

View full text Add to dashboard Cite

Training of generative models especially Generative Adversarial Networks can easily diverge in low-data setting.To mitigate this issue, we propose a novel implicit data augmentation approach which facilitates stable training and synthesize diverse samples. Specifically, we view the discriminator as a metric embedding of the real data manifold, which offers proper distances between real data points. We then utilize information in the feature space to develop a data-driven augmentation method. We further bring up a simple metric to evaluate the diversity of synthesized samples. Experiments on few-shot generation tasks show our method improves FID and diversity of results compared to current methods, and allows generating high-quality and diverse images with less than 100 training samples.

show abstract

Few-Shot Object Detection via Classification Refinement and Distractor Retreatment

Cited by 72 publications

References 11 publications

A Survey of Deep Learning for Low-Shot Object Detection

A Survey of Deep Learning for Low-Shot Object Detection

Leveraging Bottom-Up and Top-Down Attention for Few-Shot Object Detection

Implicit Data Augmentation Using Feature Interpolation for Low-Shot Image Generation

Contact Info

Product

Resources

About