What Can Be Transferred: Unsupervised Domain Adaptation for Endoscopic Lesions Segmentation

Dong, Jiahua; Cong, Yang; Sun, Gan; Zhong, Bineng; Xu, Xiaowei

doi:10.1109/cvpr42600.2020.00408

Cited by 118 publications

(39 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the future, we will consider HeDA in semantic segmentation tasks [79], [80]. Inspired by Dong et al [81], [82] that develop a novel perspective to distinguish transferable or untransferable representations across domains, we will develop a novel learning theory in semantic segmentation tasks to quantify transferability across heterogeneous domains.…”

Section: Discussionmentioning

confidence: 99%

Semi-Supervised Heterogeneous Domain Adaptation: Theory and Algorithms

Fang

Liu

et al. 2023

IEEE Trans. Pattern Anal. Mach. Intell.

View full text Add to dashboard Cite

Semi-supervised heterogeneous domain adaptation (SsHeDA) aims to train a classifier for the target domain, in which only unlabeled and a small number of labeled data are available. This is done by leveraging knowledge acquired from a heterogeneous source domain. From algorithmic perspectives, several methods have been proposed to solve the SsHeDA problem; yet there is still no theoretical foundation to explain the nature of the SsHeDA problem or to guide new and better solutions. Motivated by compatibility condition in semi-supervised probably approximately correct (PAC) theory, we explain the SsHeDA problem by proving its generalization error -that is, why labeled heterogeneous source data and unlabeled target data help to reduce the target risk. Guided by our theory, we devise two algorithms as proof of concept. One, kernel heterogeneous domain alignment (KHDA), is a kernel-based algorithm; the other, joint mean embedding alignment (JMEA), is a neural network-based algorithm. When a dataset is small, KHDA's training time is less than JMEA's. When a dataset is large, JMEA is more accurate in the target domain. Comprehensive experiments with image/text classification tasks show KHDA to be the most accurate among all non-neural network baselines, and JMEA to be the most accurate among all baselines.

show abstract

Section: Discussionmentioning

confidence: 99%

Semi-Supervised Heterogeneous Domain Adaptation: Theory and Algorithms

Fang

Liu

et al. 2023

IEEE Trans. Pattern Anal. Mach. Intell.

View full text Add to dashboard Cite

show abstract

“…With the brief mention of neural network semantic segmentation methods, including FCN [26], intensive study has been done in the computer vision society and consistently shown improved results [27][28][29]. Furthermore, with particular objectives such as semantic lesion segmentation for medical use, References [30,31] showed improved methods with highly GAN-based knowledge adaptation. However, we adopt trap ball segmentation, which is based on a supervised contour detecting line filler algorithm and the Seg2pix network, which we also use for colorization.…”

Section: Sketch Parsingmentioning

confidence: 99%

Seg2pix: Few Shot Training Line Art Colorization with Segmented Image Data

Seo

2021

Applied Sciences

View full text Add to dashboard Cite

There are various challenging issues in automating line art colorization. In this paper, we propose a GAN approach incorporating semantic segmentation image data. Our GAN-based method, named Seg2pix, can automatically generate high quality colorized images, aiming at computerizing one of the most tedious and repetitive jobs performed by coloring workers in the webtoon industry. The network structure of Seg2pix is mostly a modification of the architecture of Pix2pix, which is a convolution-based generative adversarial network for image-to-image translation. Through this method, we can generate high quality colorized images of a particular character with only a few training data. Seg2pix is designed to reproduce a segmented image, which becomes the suggestion data for line art colorization. The segmented image is automatically generated through a generative network with a line art image and a segmentation ground truth. In the next step, this generative network creates a colorized image from the line art and segmented image, which is generated from the former step of the generative network. To summarize, only one line art image is required for testing the generative model, and an original colorized image and segmented image are additionally required as the ground truth for training the model. These generations of the segmented image and colorized image proceed by an end-to-end method sharing the same loss functions. By using this method, we produce better qualitative results for automatic colorization of a particular character’s line art. This improvement can also be measured by quantitative results with Learned Perceptual Image Patch Similarity (LPIPS) comparison. We believe this may help artists exercise their creative expertise mainly in the area where computerization is not yet capable.

show abstract

“…Here, we finetune the pretrained decoder with the selected knowledge k s t by giving higher importance weights if the selected knowledge is suitable for the gold response generation. In this way, we further alleviate the mismatch problem because we highlight the matched samples by assigning an importance weight to each instance (x t , k s t , y t ) to reform the training data (Cai et al, 2020;Dong et al, 2020).…”

Section: Trainingmentioning

confidence: 99%

Unsupervised Knowledge Selection for Dialogue Generation

Chen¹,

Chen²,

Meng³

et al. 2021

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021

View full text Add to dashboard Cite

Knowledge selection is an important and challenging task which could provide the appropriate knowledge for informative dialogue generation. However, the needed gold knowledge label is difficult to collect in reality. In this paper, we study knowledge selection for dialogue generation in the unsupervised scenario and propose a novel Distilled Distant Supervision Loss (DDSL) to supervise knowledge selection when the gold knowledge label is unknown. Specifically, we first obtain an oracle knowledge label via distant supervision and then leverage knowledge distillation to alleviate the noisy labeling problem of distant supervision. Furthermore, we propose a pretraining-finetuning strategy to deal with the mismatch knowledge selection problem that models tend to select the mismatched knowledge for dialogue generation in the unsupervised setting and will cause the degeneration of knowledge-aware decoder. Experiments on two knowledge-grounded dialogue datasets show that our approach manages to select knowledge more accurately in the unsupervised setting and generates more informative responses, even outperforming many strong supervised baselines. 1

show abstract

What Can Be Transferred: Unsupervised Domain Adaptation for Endoscopic Lesions Segmentation

Cited by 118 publications

References 32 publications

Semi-Supervised Heterogeneous Domain Adaptation: Theory and Algorithms

Semi-Supervised Heterogeneous Domain Adaptation: Theory and Algorithms

Seg2pix: Few Shot Training Line Art Colorization with Segmented Image Data

Unsupervised Knowledge Selection for Dialogue Generation

Contact Info

Product

Resources

About