This paper proposes a new unsupervised domain adaptation approach called Collaborative and Adversarial Network (CAN), which uses the domain-collaborative and domain-adversarial learning for training the neural network. We show that these two learning strategies can be uniformly formulated as a domain classifier learning with positive or negative weights on the losses. Moreover, to further enhance the descriminability in the target domain, we propose Self-Paced CAN (SPCAN), which progressively selects pseudo-labeled target samples for re-training the classifiers. Additionally, we extend our domain adaptation approach for more difficult video action recognition task, which is based on the popular two stream approach but additionally considering the cooperation between the RGB stream and the optical flow stream. We propose the Cooperative SPCAN (CoSPCAN) method to select and reweigh the pseudo labeled target samples of one stream (RGB/Flow) based on the information from another stream (Flow/RGB) in a cooperative way. As a result, our CoSPCAN model is able to exchange the information between the two streams. Comprehensive experiments on different benchmark datasets for the object recognition task and the video action recognition task, show our newly proposed approach achieves state-of-the-art performance, which clearly demonstrate the effectiveness of our proposed approaches for unsupervised domain adaptation.

In this work, we propose a new generic multi-modality domain adaptation framework called Progressive Modality Cooperation (PMC) to transfer the knowledge learned from the source domain to the target domain by exploiting multiple modality clues ( e.g. , RGB and depth) under the multi-modality domain adaptation (MMDA) and the more general multi-modality domain adaptation using privileged information (MMDA-PI) settings. Under the MMDA setting, the samples in both domains have all the modalities. Through effective collaboration among multiple modalities, the two newly proposed modules in our PMC can select the reliable pseudo-labeled target samples, which captures the modality-specific information and modality-integrated information, respectively. Under the MMDA-PI setting, some modalities are missing in the target domain. Hence, to better exploit the multi-modality data in the source domain, we further propose the PMC with privileged information (PMC-PI) method by proposing a new multi-modality data generation (MMG) network. MMG generates the missing modalities in the target domain based on the source domain data by considering both domain distribution mismatch and semantics preservation, which are respectively achieved by using adversarial learning and conditioning on weighted pseudo semantic class labels. Extensive experiments on three image datasets and eight video datasets for various multi-modality cross-domain visual recognition tasks under both MMDA and MMDA-PI settings clearly demonstrate the effectiveness of our proposed PMC framework.

Background: Small nonfunctional pancreatic neuroendocrine tumors (NF-PNETs) ≤2 cm have variable biological features, and there is no gold standard treatment for their management. The present study aimed to evaluate the risk of malignancy of small NF-PNETs and their outcomes following curative resection. Methods: Patients with NF-PNETs undergoing surgical resection at the First Affiliated Hospital, College of Medicine, Zhejiang University, between 2012 and 2017 were included. Clinicopathological characteristics, perioperative results, and prognosis were retrospectively analyzed. Results: A total of 73 patients were identified, including 28 with small NF-PNETs and 45 large PNETs; 32.1% of NF-PNETs ≤2 cm underwent a parenchyma-sparing pancreas surgery, which was >6.7% in large NF-PNETs. No statistically significant differences in perioperative results, postoperative complications, and long-term outcomes were found between small tumors undergoing standard and parenchyma-sparing pancreatectomy. Eighteen small tumors (64.3%) developed a perioperative complication, with a clinically significant pancreatic fistula rate of 25%; however, only 2 patient needed reintervention. Small NF-PNETs in 3 patients were malignant. Multivariate logistic regression showed that grade ≥3 and lymphovascular invasion were independently related to malignancy in NF-PNETs. Conclusions: Small NF-PNETs (≤2 cm) are not immune from potential malignancy. Surgical resection may be considered for small tumors and can provide favorable postoperative and long-term outcomes. Parenchyma-sparing pancreatectomy may be an alternative surgery for selected small local NF-PNETs.

The rapid development of UAV-LiDAR and data processing capabilities is likely to enable accurate individual-tree inventories in the near future, requiring few on-ground calibration measurements. Using data collected from 20 radiata pine trials dispersed across New Zealand, the objective of this study was to determine the accuracy of high-density UAV-LiDAR for the prediction of tree diameter and volume, under a range of data calibration scenarios. Using all measurements for the calibration (a range of 335–4703 tree measurements across the 20 sites), accurate random forest models for each of the 20 sites were created from a diverse range of LiDAR metrics that characterised the horizontal and vertical structures of the canopy. Averaged across the 20 sites, predictions had a mean R² and relative RMSE (rRMSE) of, respectively, 0.713 and 9.699% for the tree diameter and 0.746 and 19.57% for the tree volume. Reductions in the numbers of calibration trees per trial had little effect on model accuracy until only 300 trees/site were used; however, accurate, unbiased predictions were still possible using as few as 100 trees/site. More generally, applicable random forest models for both tree dimensions were constructed by collating all of the data and tested using leave-one-site-out cross-validation to determine the accuracy of the model predictions when calibration measurements were not available. The predictions using this approach were reasonable but less accurate and more biased than with the use of calibration data, with a mean R² and rRMSE of, respectively, 0.631 and 15.12% for the tree diameter and 0.631 and 35.6% for the volume. Our research aims to facilitate the transition from a plot-based to tree-level inventory in plantation forests and contribute to the future development of a generalised model that could accurately predict tree dimensions from UAV-LiDAR, relying on minimal field measurements.