Publications by authors named "Junshi Huang"
Contrastive Open-Set Active Learning-Based Sample Selection for Image Classification.
IEEE Trans Image Process
October 2024
Article Synopsis
- The paper discusses a practical challenge in Active Learning (AL) called open-set AL, where unlabeled data includes both in-distribution (ID) and out-of-distribution (OOD) samples, causing standard methods to underperform.
- The authors propose two new criteria, contrastive confidence and historical divergence, to effectively select more informative ID samples while minimizing the selection of less helpful OOD samples.
- They introduce a contrastive clustering framework that allows the classifier to better recognize OOD samples and improve overall representation learning, achieving state-of-the-art results on various benchmark datasets.
Unpaired Image-text Matching via Multimodal Aligned Conceptual Knowledge.
IEEE Trans Pattern Anal Mach Intell
July 2024
Article Synopsis
- Recent advances in multimodal pretrained models have improved image-text matching using vast datasets of paired images and texts, but these models rely heavily on supervised learning.
- This paper introduces a novel approach called Multimodal Aligned Conceptual Knowledge (MACK) for unpaired image-text matching, where paired data is not available during training.
- MACK involves collecting and refining knowledge from unpaired datasets through self-supervised learning, allowing for the computation of image-text similarity scores, and can enhance existing models' performance, particularly in zero-shot and cross-dataset scenarios.
Article Synopsis
- Surface-enhanced Raman spectroscopy (SERS) was used to analyze chlorpyrifos pesticide residues in rice, using gold nanoparticles to boost signal strength and various chemicals for purification.
- A successive projections algorithm (SPA) helped identify key Raman peaks related to the pesticide, leading to efficient quantitative analysis using support vector machine (SVM) and partial least squares (PLS) models.
- The study showed that analyzing only six selected Raman peaks resulted in high accuracy and a quick analysis time of 10 minutes, with excellent recovery rates for unknown samples, indicating SERS is effective for monitoring pesticide residues in rice.
Broad-spectrum resistance has great values for crop breeding. However, its mechanisms are largely unknown. Here, we report the cloning of a maize NLR gene, RppK, for resistance against southern corn rust (SCR) and its cognate Avr gene, AvrRppK, from Puccinia polysora (the causal pathogen of SCR).
View Article and Find Full Text PDFNatural alleles that control multiple disease resistance (MDR) are valuable for crop breeding. However, only one MDR gene has been cloned in maize, and the molecular mechanisms of MDR remain unclear in maize. In this study, through map-based cloning we cloned a teosinte-derived allele of a resistance gene, Mexicana lesion mimic 1 (ZmMM1), which causes a lesion mimic phenotype and confers resistance to northern leaf blight (NLB), gray leaf spot (GLS), and southern corn rust (SCR) in maize.
View Article and Find Full Text PDFTowards Robust and Accurate Multi-View and Partially-Occluded Face Alignment.
IEEE Trans Pattern Anal Mach Intell
April 2018
Face alignment acts as an important task in computer vision. Regression-based methods currently dominate the approach to solving this problem, which generally employ a series of mapping functions from the face appearance to iteratively update the face shape hypothesis. One keypoint here is thus how to perform the regression procedure.
View Article and Find Full Text PDFHCP: A Flexible CNN Framework for Multi-label Image Classification.
IEEE Trans Pattern Anal Mach Intell
September 2016
Convolutional Neural Network (CNN) has demonstrated promising performance in single-label image classification tasks. However, how CNN best copes with multi-label images still remains an open problem, mainly due to the complex underlying object layouts and insufficient multi-label training images. In this work, we propose a flexible deep CNN infrastructure, called Hypotheses-CNN-Pooling (HCP), where an arbitrary number of object segment hypotheses are taken as the inputs, then a shared CNN is connected with each hypothesis, and finally the CNN output results from different hypotheses are aggregated with max pooling to produce the ultimate multi-label predictions.
View Article and Find Full Text PDF