Training deep retrieval models with noisy datasets: Bag exponential loss

Martínez-Cortés, Tomás; González-Díaz, Iván; Díaz-de-María, Fernando

doi:10.1016/j.patcog.2020.107811

Cited by 5 publications

(4 citation statements)

References 121 publications

(197 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In [10] vision transformers are proposed to generate image descriptors for retrieval, training the models with a metric learning objective that combines a contrastive loss with a differential entropy regularizer. In [26], the problem of retrieval under noisy datasets is addressed proposing a novel noiserobust loss based on Multiple Instance Learning (MIL). The proposed method allows to use noisy generated training sets, that can be easily created, to adapt CNNs for image retrieval on new objects.…”

Section: End-to-end Approachesmentioning

confidence: 99%

Effective triplet mining improves training of multi-scale pooled CNN for image retrieval

Vaccaro

Bertini

Uricchio

et al. 2022

Machine Vision and Applications

View full text Add to dashboard Cite

In this paper, we address the problem of content-based image retrieval (CBIR) by learning images representations based on the activations of a Convolutional Neural Network. We propose an end-to-end trainable network architecture that exploits a novel multi-scale local pooling based on the trainable aggregation layer NetVLAD (Arandjelovic et al in Proceedings of the IEEE conference on computer vision and pattern recognition CVPR, NetVLAD, 2016) and bags of local features obtained by splitting the activations, allowing to reduce the dimensionality of the descriptor and to increase the performance of retrieval. Training is performed using an improved triplet mining procedure that selects samples based on their difficulty to obtain an effective image representation, reducing the risk of overfitting and loss of generalization. Extensive experiments show that our approach, that can be effectively used with different CNN architectures, obtains state-of-the-art results on standard and challenging CBIR datasets.

show abstract

Section: End-to-end Approachesmentioning

confidence: 99%

Effective triplet mining improves training of multi-scale pooled CNN for image retrieval

Vaccaro

Bertini

Uricchio

et al. 2022

Machine Vision and Applications

View full text Add to dashboard Cite

show abstract

“…In specific projects, some loss functions decrease the gradient of the difference calculated by the gradient fast, while some decrease slowly, so choosing the appropriate loss function is also very critical. Currently, the commonly used loss functions in the training of classification neural networks include 0-1 loss, Logistics loss, Hinge loss [29], exponential loss [30], and cross-entropy loss [31]. In this paper, the loss function we use is the cross-entropy loss function.…”

Section: Loss Functionmentioning

confidence: 99%

Optimizing the performance of the neural network by using a mini dataset processing method

chen

2022

Preprint

View full text Add to dashboard Cite

Data processing is one of the essential methods to optimize the performance of neural networks. In this paper, we give up the traditional data processing method and propose a method to optimize the deep neural network by processing the mini data set based on the loss. Using this method, each Literation training can obtain a relatively optimal result, and the optimization effects of each time are integrated to optimize the results of each epoch finally At the same time, in order to verify the effectiveness of this data processing method, experiments are carried out on MNIST, HAGRID, and CIFAR-10 datasets to compare the effects of using this method and not using this method under different hyperparameters, and finally, the effectiveness of this data processing method is verified. On this basis, the advantages and disadvantages of this method are summarized. Finally, the future research direction has been prospected.

show abstract

“…This is different from the traditional triplet mining methods that often address the negative samples (i.e., heterogeneous pairs). It should be mentioned that the idea of addressing the importance of similar homogeneous pairs has been introduced by BagLoss [33] to improve the performance of image retrieval models. The technical difference between BagLoss and GraVIS lies in the way to generate and utilize negative samples.…”

Section: Related Workmentioning

confidence: 99%

“…For contrastive approaches, we include SimCLR-Derm [3], MoCov2 [43], BYOL [21] and C2L [2] for comparison. As for traditional self-supervised learning methods, we compare GraVIS against Model Genesis (MG) [1], N -pairs triplet loss with the triplet mining strategy (NPTL) [44] and BagLoss [33], where the last two methods are based on metric learning.…”

Section: Baselinesmentioning

confidence: 99%

GraVIS: Grouping Augmented Views From Independent Sources for Dermatology Analysis

Lu¹,

Wang²,

Yu³

2022

IEEE Trans. Med. Imaging

View full text Add to dashboard Cite

Self-supervised representation learning has been extremely successful in medical image analysis, as it requires no human annotations to provide transferable representations for downstream tasks. Recent self-supervised learning methods are dominated by noise-contrastive estimation (NCE, also known as contrastive learning), which aims to learn invariant visual representations by contrasting one homogeneous image pair with a large number of heterogeneous image pairs in each training step. Nonetheless, NCE-based approaches still suffer from one major problem that is one homogeneous pair is not enough to extract robust and invariant semantic information. Inspired by the archetypical triplet loss, we propose GraVIS, which is specifically optimized for learning self-supervised features from dermatology images, to group homogeneous dermatology images while separating heterogeneous ones. In addition, a hardness-aware attention is introduced and incorporated to address the importance of homogeneous image views with similar appearance instead of those dissimilar homogeneous ones. GraVIS significantly outperforms its transfer learning and self-supervised learning counterparts in both lesion segmentation and disease classification tasks, sometimes by 5 percents under extremely limited supervision. More importantly, when equipped with the pretrained weights provided by GraVIS, a single model could achieve better results than winners that heavily rely on ensemble strategies in the well-known ISIC 2017 challenge. Code is available at https://bit.ly/3xiFyjx.

show abstract

Training deep retrieval models with noisy datasets: Bag exponential loss

Abstract: Temporal segmentation and keyframe selection methods for user-generated video search-based annotation. Expert Systems with Applications, 42(1), 488502.

Cited by 5 publications

References 121 publications

Effective triplet mining improves training of multi-scale pooled CNN for image retrieval

Effective triplet mining improves training of multi-scale pooled CNN for image retrieval

Optimizing the performance of the neural network by using a mini dataset processing method

GraVIS: Grouping Augmented Views From Independent Sources for Dermatology Analysis

Contact Info

Product

Resources

About