The class activation maps are generated from the final convolutional layer of CNN. They can highlight discriminative object regions for the class of interest. These discovered object regions have ...been widely used for weakly-supervised tasks. However, due to the small spatial resolution of the final convolutional layer, such class activation maps often locate coarse regions of the target objects, limiting the performance of weakly-supervised tasks that need pixel-accurate object locations. Thus, we aim to generate more fine-grained object localization information from the class activation maps to locate the target objects more accurately. In this paper, by rethinking the relationships between the feature maps and their corresponding gradients, we propose a simple yet effective method, called LayerCAM. It can produce reliable class activation maps for different layers of CNN. This property enables us to collect object localization information from coarse (rough spatial localization) to fine (precise fine-grained details) levels. We further integrate them into a high-quality class activation map, where the object-related pixels can be better highlighted. To evaluate the quality of the class activation maps produced by LayerCAM, we apply them to weakly-supervised object localization and semantic segmentation. Experiments demonstrate that the class activation maps generated by our method are more effective and reliable than those by the existing attention methods. The code will be made publicly available.
Delving Deep Into Label Smoothing Zhang, Chang-Bin; Jiang, Peng-Tao; Hou, Qibin ...
IEEE transactions on image processing,
2021, Letnik:
30
Journal Article
Recenzirano
Label smoothing is an effective regularization tool for deep neural networks (DNNs), which generates soft labels by applying a weighted average between the uniform distribution and the hard label. It ...is often used to reduce the overfitting problem of training DNNs and further improve classification performance. In this paper, we aim to investigate how to generate more reliable soft labels. We present an Online Label Smoothing (OLS) strategy, which generates soft labels based on the statistics of the model prediction for the target category. The proposed OLS constructs a more reasonable probability distribution between the target categories and non-target categories to supervise DNNs. Experiments demonstrate that based on the same classification models, the proposed approach can effectively improve the classification performance on CIFAR-100, ImageNet, and fine-grained datasets. Additionally, the proposed method can significantly improve the robustness of DNN models to noisy labels compared to current label smoothing approaches. The source code is available at our project page: https://mmcheng.net/ols/
Deep Hough Transform for Semantic Line Detection Zhao, Kai; Han, Qi; Zhang, Chang-Bin ...
IEEE transactions on pattern analysis and machine intelligence,
09/2022, Letnik:
44, Številka:
9
Journal Article
Recenzirano
We focus on a fundamental task of detecting meaningful line structures, a.k.a. , semantic line, in natural scenes. Many previous methods regard this problem as a special case of object detection and ...adjust existing object detectors for semantic line detection. However, these methods neglect the inherent characteristics of lines, leading to sub-optimal performance. Lines enjoy much simpler geometric property than complex objects and thus can be compactly parameterized by a few arguments. To better exploit the property of lines, in this paper, we incorporate the classical Hough transform technique into deeply learned representations and propose a one-shot end-to-end learning framework for line detection. By parameterizing lines with slopes and biases, we perform Hough transform to translate deep representations into the parametric domain, in which we perform line detection. Specifically, we aggregate features along candidate lines on the feature map plane and then assign the aggregated features to corresponding locations in the parametric domain. Consequently, the problem of detecting semantic lines in the spatial domain is transformed into spotting individual points in the parametric domain, making the post-processing steps, i.e., non-maximal suppression, more efficient. Furthermore, our method makes it easy to extract contextual line features that are critical for accurate line detection. In addition to the proposed method, we design an evaluation metric to assess the quality of line detection and construct a large scale dataset for the line detection task. Experimental results on our proposed dataset and another public dataset demonstrate the advantages of our method over previous state-of-the-art alternatives. The dataset and source code is available at https://mmcheng.net/dhtline/ .
Reducing the size of heterogeneous nanocatalysts is generally conducive to improving their atomic utilization and activities in various catalytic reactions. However, this strategy has proven less ...effective for Cu-based electrocatalysts for the reduction of CO
2
to multicarbon (C
2+
) products, owing to the overly strong binding of intermediates on small-sized (< 15 nm) Cu nanoparticles (NPs). Herein, by incorporating pyrenyl-graphdiyne (Pyr-GDY), we successfully endowed ultrafine (∼ 2 nm) Cu NPs with a significantly elevated selectivity for CO
2
-to-C
2+
conversion. The Pyr-GDY can not only help to relax the overly strong binding between adsorbed H* and CO* intermediates on Cu NPs by tailoring the d-band center of the catalyst, but also stabilize the ultrafine Cu NPs through the high affinity between alkyne moieties and Cu NPs. The resulting Pyr-GDY-Cu composite catalyst gave a Faradic efficiency (FE) for C
2+
products up to 74%, significantly higher than those of support-free Cu NPs (C
2+
FE, ~ 2%), carbon nanotube-supported Cu NPs (CNT-Cu, C
2+
FE, ~ 18%), graphene oxide-supported Cu NPs (GO-Cu, C
2+
FE, ~ 8%), and other reported ultrafine Cu NPs. Our results demonstrate the critical influence of graphdiyne on the selectivity of Cu-catalyzed CO
2
electroreduction, and showcase the prospect for ultrafine Cu NPs catalysts to convert CO
2
into value-added C
2+
products.
Constituting approximately 10% of flowering plant species, orchids (Orchidaceae) display unique flower morphologies, possess an extraordinary diversity in lifestyle, and have successfully colonized ...almost every habitat on Earth. Here we report the draft genome sequence of Apostasia shenzhenica, a representative of one of two genera that form a sister lineage to the rest of the Orchidaceae, providing a reference for inferring the genome content and structure of the most recent common ancestor of all extant orchids and improving our understanding of their origins and evolution. In addition, we present transcriptome data for representatives of Vanilloideae, Cypripedioideae and Orchidoideae, and novel third-generation genome data for two species of Epidendroideae, covering all five orchid subfamilies. A. shenzhenica shows clear evidence of a whole-genome duplication, which is shared by all orchids and occurred shortly before their divergence. Comparisons between A. shenzhenica and other orchids and angiosperms also permitted the reconstruction of an ancestral orchid gene toolkit. We identify new gene families, gene family expansions and contractions, and changes within MADS-box gene classes, which control a diverse suite of developmental processes, during orchid evolution. This study sheds new light on the genetic mechanisms underpinning key orchid innovations, including the development of the labellum and gynostemium, pollinia, and seeds without endosperm, as well as the evolution of epiphytism; reveals relationships between the Orchidaceae subfamilies; and helps clarify the evolutionary history of orchids within the angiosperms.
Display omitted
•Plate-shaped CoMn2O4 was synthesized using MOFs as precursor.•The catalyst showed large specific surface area and abundant active sites.•It had much higher catalytic activity than ...previously-reported CoMn2O4 catalyst.•The Co-Mn synergy in heterogeneous catalysis were elucidated.•The material showed good stability and reusability for SA removal.
Spinel-type CoMn2O4 materials are promising catalyst for heterogeneous activation of peroxymonosulfate (PMS), but the catalytic activity still need considerable improvements for practical environmental application and the underlying Co-Mn synergy is unclear. In this work, we synthesized CoMn2O4 microplates by using CoMn2-perylene-3,4,9,10-tetracarboxylic dianhydride (ptcda) metal organic frameworks (MOFs) as the precursor. The resulting material showed significantly higher catalytic activity for the PMS activation and sulfanilamide (SA) degradation than the CoMn2O4 obtained by conventional solvothermal synthesis methods, due to its much higher specific surface area and abundant surface hydroxyl groups as the active sites. In addition, the Co-Mn synergy in the synthesized material for the efficient heterogeneous catalysis was elucidated. The catalyst stability was also evaluated. Our work may lay the foundation for optimized design of highly-efficient heterogeneous catalyst for environmental application.
Carbon nanotubes (CNTs) have attracted great attentions in the field of electronics, sensors, healthcare, and energy conversion. Such emerging applications have driven the carbon nanotube research in ...a rapid fashion. Indeed, the structure control over CNTs has inspired an intensive research vortex due to the high promises in electronic and optical device applications. Here, this in-depth review is anticipated to provide insights into the controllable synthesis and applications of high-quality CNTs. First, the general synthesis and post-purification of CNTs are briefly discussed. Then, the state-of-the-art electronic device applications are discussed, including field-effect transistors, gas sensors, DNA biosensors, and pressure gauges. Besides, the optical sensors are delivered based on the photoluminescence. In addition, energy applications of CNTs are discussed such as thermoelectric energy generators. Eventually, future opportunities are proposed for the Internet of Things (IoT) oriented sensors, data processing, and artificial intelligence.
Understanding the genetic changes underlying phenotypic variation in sheep (Ovis aries) may facilitate our efforts towards further improvement. Here, we report the deep resequencing of 248 sheep ...including the wild ancestor (O. orientalis), landraces, and improved breeds. We explored the sheep variome and selection signatures. We detected genomic regions harboring genes associated with distinct morphological and agronomic traits, which may be past and potential future targets of domestication, breeding, and selection. Furthermore, we found non-synonymous mutations in a set of plausible candidate genes and significant differences in their allele frequency distributions across breeds. We identified PDGFD as a likely causal gene for fat deposition in the tails of sheep through transcriptome, RT-PCR, qPCR, and Western blot analyses. Our results provide insights into the demographic history of sheep and a valuable genomic resource for future genetic studies and improved genome-assisted breeding of sheep and other domestic animals.
Elymus breviaristatus and Elymus sinosubmuticus are perennial herbs, not only morphologically similar but also sympatric distribution. The genome composition of E. sinosubmuticus has not been ...reported, and the relationship between E. sinosubmuticus and E. breviaristatus is still controversial. We performed artificial hybridization, genomic in situ hybridization, and phylogenetic analyses to clarify whether the two taxa were the same species.
The high frequency bivalent (with an average of 20.62 bivalents per cell) at metaphase I of pollen mother cells of the artificial hybrids of E. breviaristatus (StYH) × E. sinosubmuticus was observed. It illustrated that E. sinosubmuticus was closely related to E. breviaristatus. Based on genomic in situ hybridization results, we confirmed that E. sinosubmuticus was an allohexaploid, and the genomic constitution was StYH. Phylogenetic analysis results also supported that this species contained St, Y, and H genomes. In their F
hybrids, pollen activity was 53.90%, and the seed setting rate was 22.46%. Those indicated that the relationship between E. sinosubmuticus and E. breviaristatus is intersubspecific rather than interspecific, and it is reasonable to treated E. sinosubmuticus as the subspecies of E. breviaristatus.
In all, the genomic constitutions of E. sinosubmuticus and E. breviaristatus were StYH, and they are species in the genus Campeiostachys. Because E. breviaristatus was treated as Campeistachys breviaristata, Elymus sinosubmuticus should be renamed Campeiostachys breviaristata (Keng) Y. H. Zhou, H. Q. Zhang et C. R. Yang subsp. sinosubmuticus (S. L. Chen) Y. H. Zhou, H. Q. Zhang et L. Tan.
Replication fork reversal which restrains DNA replication progression is an important protective mechanism in response to replication stress. PARP1 is recruited to stalled forks to restrain DNA ...replication. However, PARP1 has no helicase activity, and the mechanism through which PARP1 participates in DNA replication restraint remains unclear. Here, we found novel protein-protein interactions between PARP1 and DNA translocases, including HLTF, SHPRH, ZRANB3, and SMARCAL1, with HLTF showing the strongest interaction among these DNA translocases. Although HLTF and SHPRH share structural and functional similarity, it remains unclear whether SHPRH contains DNA translocase activity. We further identified the ability of SHPRH to restrain DNA replication upon replication stress, indicating that SHPRH itself could be a DNA translocase or a helper to facilitate DNA translocation. Although hydroxyurea (HU) and MMS induce different types of replication stress, they both induce common DNA replication restraint mechanisms independent of intra-S phase activation. Our results suggest that the PARP1 facilitates DNA translocase recruitment to damaged forks, preventing fork collapse and facilitating DNA repair.