Search

Abstract Genome variant calling is a challenging yet critical task for subsequent studies. Existing methods almost rely on high depth DNA sequencing data. Performance on low depth data drops a lot. Using public Oxford Nanopore (ONT) data of human being from the Genome in a Bottle (GIAB) Consortium, we trained a generative adversarial network for low depth variant calling. Our method, noted as LDV-Caller, can project high depth sequencing information from low depth data. It achieves 94.25% F1 score on low depth data, while the F1 score of the state-of-the-art method on two times higher depth data is 94.49%. By doing so, the price of genome-wide sequencing examination can reduce deeply. In addition, we validated the trained LDV-Caller model on 157 public Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) samples. The mean sequencing depth of these samples is 2982. The LDV-Caller yields 92.77% F1 score using only 22x sequencing depth, which demonstrates our method has potential to analyze different species with only low depth sequencing data.

Intra-camera supervision (ICS) for person re-identification (Re-ID) assumes that identity labels are independently annotated within each camera view and no inter-camera identity association is labeled. It is a new setting proposed recently to reduce the burden of annotation while expect to maintain desirable Re-ID performance. However, the lack of inter-camera labels makes the ICS Re-ID problem much more challenging than the fully supervised counterpart. By investigating the characteristics of ICS, this article proposes a graph-induced contrastive learning (GCL) approach to address this issue. More specifically, we first formulate the cross-camera ID association task as a graph partitioning problem subjected to ICS-specific constraints and design a greedy agglomeration algorithm to solve it. Then, we propose a graph-induced contrastive loss that unifies both intra- and inter-camera learning into a contrastive learning framework to learn a Re-ID model. The cross-camera ID association step and the Re-ID model contrastive learning step are alternatively iterated, by which we progressively obtain a highly discriminative Re-ID model. Extensive experiments on three large-scale datasets show that our approach outperforms all previous ICS works. Especially, it gains 15.7% Rank-1 and 14.3% mAP improvements on the challenging MSMT17 dataset. Moreover, our approach performs even comparable to state-of-the-art fully supervised methods on all of the three datasets.

Forecasting future traffic flows from previous ones is a challenging problem because of their complex and dynamic nature of spatio-temporal structures. Most existing graph-based CNNs attempt to capture the static relations while largely neglecting the dynamics underlying sequential data. In this article, we present dynamic spatio-temporal graph-based CNNs (DST-GCNNs) by learning expressive features to represent spatio-temporal structures and predict future traffic flows from surveillance video data. In particular, DST-GCNN is a two stream network. In the flow prediction stream, we present a novel graph-based spatio-temporal convolutional layer to extract features from a graph representation of traffic flows. Then several such layers are stacked together to predict future flows over time. Meanwhile, the relations between traffic flows in the graph are often time variant as the traffic condition changes over time. To capture the graph dynamics, we use the graph prediction stream to predict the dynamic graph structures, and the predicted structures are fed into the flow prediction stream. Experiments on real datasets demonstrate that the proposed model achieves competitive performances compared with the other state-of-the-art methods.

We exploit the criteria to optimize training set construction for the large-scale video semantic classification. Due to the large gap between low-level features and higher-level semantics, as well as the high diversity of video data, it is difficult to represent the prototypes of semantic concepts by a training set of limited size. In video semantic classification, most of the learning-based approaches require a large training set to achieve good generalization capacity, in which large amounts of labor-intensive manual labeling are ineluctable. However, it is observed that the generalization capacity of a classifier highly depends on the geometrical distribution of the training data rather than the size. We argue that a training set which includes most temporal and spatial distribution information of the whole data will achieve a good performance even if the size of training set is limited. In order to capture the geometrical distribution characteristics of a given video collection, we propose four metrics for constructing/selecting an optimal training set, including salience, temporal dispersiveness, spatial dispersiveness, and diversity. Furthermore, based on these metrics, we propose a set of optimization rules to capture the most distribution information of the whole data using a training set with a given size. Experimental results demonstrate these rules are effective for training set construction in video semantic classification, and significantly outperform random training set selection.