Learning From Web Data With Self-Organizing Memory Module

Tu, Yi; Niu, Li; Chen, Junjie; Cheng, Dawei; Zhang, Liqing

doi:10.1109/cvpr42600.2020.01286

Cited by 17 publications

(12 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the past few years, learning from web data has received widespread attention in the vision community [7], [15], [18]- [21] and led to great success in a variety of tasks including scene classification [19], clothing recognition [7], skin disease diagnosis [15], and action recognition [22]- [25], etc. Several works [10], [15], [26] gather web images from common web search engines for given query classes.…”

Section: A Recognition From Web Datamentioning

confidence: 99%

BiSPL: Bidirectional Self-Paced Learning for Recognition From Web Data

Chang

Lai

et al. 2021

IEEE Trans. on Image Process.

View full text Add to dashboard Cite

Deep learning (DL) is inherently subject to the requirement of a large amount of well-labeled data, which is expensive and time-consuming to obtain manually. In order to broaden the reach of DL, leveraging free web data becomes an attractive strategy to alleviate the issue of data scarcity. However, directly utilizing collected web data to train a deep model is ineffective because of the mixed noisy data. To address such problems, we develop a novel bidirectional self-paced learning (BiSPL) framework which reduces the effect of noise by learning from web data in a meaningful order. Technically, the BiSPL framework consists of two essential steps. Relying on distances defined between web samples and labeled source samples, first, the web samples with short distances are sampled and combined to form a new training set. Second, based on the new training set, both easy and hard samples are initially employed to train deep models for higher stability, and hard samples are gradually dropped to reduce the noise as the training progresses. By iteratively alternating such steps, deep models converge to a better solution. We mainly focus on the fine-grained visual classification (FGVC) tasks because their corresponding datasets are generally small and therefore face a more significant data scarcity problem. Experiments conducted on six public FGVC tasks demonstrate that our proposed method outperforms the state-of-the-art approaches. Especially, BiSPL suffices to achieve the highest stable performance when the scale of the well-labeled training set decreases dramatically.

show abstract

Section: A Recognition From Web Datamentioning

confidence: 99%

BiSPL: Bidirectional Self-Paced Learning for Recognition From Web Data

Chang

Lai

et al. 2021

IEEE Trans. on Image Process.

View full text Add to dashboard Cite

show abstract

“…The feature maps from each up-sampling are concatenated with the feature maps output from the corresponding layer on the left. The concatenated feature map is then fed into the next decode block [12]. Nucleus masks of the same size as the input image are finally detected.…”

Section: The Nucleus Segmentationmentioning

confidence: 99%

Nuclear Density Distribution Feature for Improving Cervical Histopathological Images Recognition

Wang

Yang

Lyu

et al. 2021

2021 IEEE International Conference on Image Processing (ICIP)

View full text Add to dashboard Cite

Cervical carcinoma is a common type of cancer in the female reproductive system. Early detection and diagnosis can facilitate immediate treatment and prevent progression of the disease. However, in order to achieve better performance, DL-based algorithms just stack various layers with low interpretability. In this paper, a robust and reliable Nuclear Density Distribution Feature (NDDF) based on priors of the pathologists to promote the Cervical Histopathological Image Classification (CHIC) is proposed. Our proposed method combines the nucleus mask segmented by U-Net with the segmentation grid-lines generated from pathology images utilizing SLIC to obtain the NDDF map, which contains information about the morphology, size, number, and spatial distribution of nuclei. The result shows that the proposed model trained with NDDF maps has better performance and accuracy than that trained on RGB images (patch-level histopathological images). More significantly, the accuracy of the twostream network trained with RGB images and NDDF maps is steadily improved over the corresponding baselines of different complexity.

show abstract

“…Rather than our normally used ResNet50-D, we use standard ResNet50 here for fair comparisons with others. While all the CleanNet [17] 83.95 Guidance Learning [18] 84.20 MetaCleaner [40] 85.05 Deep Self-Learning [9] 85.11 SOMNet [34] 87 other methods (except [40]) train Food-101N from ImageNet pretrained model, we train our model from scratch and still reach optimal performance.…”

Section: Ablation Studymentioning

confidence: 99%

Webly Supervised Image Classification with Self-contained Confidence

Yang

Feng

Chen

et al. 2020

Lecture Notes in Computer Science

View full text Add to dashboard Cite

This paper focuses on webly supervised learning (WSL), where datasets are built by crawling samples from the Internet and directly using search queries as web labels. Although WSL benefits from fast and low-cost data collection, noises in web labels hinder better performance of the image classification model. To alleviate this problem, in recent works, self-label supervised loss Ls is utilized together with webly supervised loss Lw. Ls relies on pseudo labels predicted by the model itself. Since the correctness of the web label or pseudo label is usually on a case-by-case basis for each web sample, it is desirable to adjust the balance between Ls and Lw on sample level. Inspired by the ability of Deep Neural Networks (DNNs) in confidence prediction, we introduce Self-Contained Confidence (SCC) by adapting model uncertainty for WSL setting, and use it to sample-wisely balance Ls and Lw. Therefore, a simple yet effective WSL framework is proposed. A series of SCC-friendly regularization approaches are investigated, among which the proposed graph-enhanced mixup is the most effective method to provide high-quality confidence to enhance our framework. The proposed WSL framework has achieved the state-of-the-art results on two largescale WSL datasets, WebVision-1000 and Food101-N. Code is available at https://github.com/bigvideoresearch/SCC.

show abstract

Learning From Web Data With Self-Organizing Memory Module

Cited by 17 publications

References 39 publications

BiSPL: Bidirectional Self-Paced Learning for Recognition From Web Data

BiSPL: Bidirectional Self-Paced Learning for Recognition From Web Data

Nuclear Density Distribution Feature for Improving Cervical Histopathological Images Recognition

Webly Supervised Image Classification with Self-contained Confidence

Contact Info

Product

Resources

About