Fine-grained Patches Research Articles

ECTFormer: An efficient Conv-Transformer model design for image recognition

Since the success of Vision Transformers (ViTs), there has been growing interest in combining ConvNets and Transformers in the computer vision community. While the hybrid models have demonstrated state-of-the-art performance, many of these models are too large and complex to be applied to edge devices for real-world applications. To address this challenge, we propose an efficient hybrid network called ECTFormer that leverages the strengths of ConvNets and Transformers while considering both model performance and inference speed. Specifically, our approach involves: (1) optimizing the combination of convolution kernels by dynamically adjusting kernel sizes based on the scale of feature tensors; (2) revisiting existing overlapping patchify to not only reduce the model size but also propagate fine-grained patches for the performance enhancement; and (3) introducing an efficient single-head self-attention mechanism, rather than multi-head self-attention in the base Transformer, to minimize the increase in model size and boost inference speed, overcoming bottlenecks of ViTs. In experimental results on ImageNet-1K, ECTFormer not only demonstrates comparable or higher top-1 accuracy but also faster inference speed on both GPUs and edge devices compared to other efficient networks.

Fine-Grained Patch Segmentation and Rasterization for 3-D Point Cloud Attribute Compression

Due to the high dimensionality of point cloud data and the irregularity and complexity of its geometric structure, effective attribute compression remains a very challenging task. Many recent efforts have focused on transforming point clouds into images and leveraging existing sophisticated image/video codecs to improve attribute coding efficiency. However, how to synthesize coherent and correlation-preserving attribute images is still inadequately addressed by existing studies, which are hindering the exertion of the merits of well-developed compression infrastructure. In this paper, we present a novel image synthesis method for effective point cloud attribute compression. Firstly, the proposed scheme segments a given point cloud into a collection of fine-grained patches by performing geometric structure analysis using heat kernel signature feature descriptor and complex points; Secondly, we transform the obtained patches from 3-D to 2-D using a low-dimensional embedding algorithm and then convert them into patch attribute images with the proposed patch rasterization and rectification method; And finally, we compactly assemble all the attribute images of patches together by formulating it as a bin nesting problem and harvest an attribute image of the whole point cloud for image/video-based compression. Experimental results demonstrate the effectiveness of the proposed method in point cloud attribute compression and its superiority over state-of-the-art codecs. The source code of this work is publicly available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/pccompession/UPCAC</uri> .

MAVA: Multi-level Adaptive Visual-textual Alignment by Cross-media Bi-attention Mechanism.

The rapidly developing information technology leads to a fast growth of visual and textual contents, and it comes with huge challenges to make correlation and perform crossmedia retrieval between images and sentences. Existing methods mainly explore cross-media correlation from either global-level instances as the whole images and sentences, or local-level fine-grained patches as the discriminative image regions and key words, which ignore the complementary information from the relation between local-level fine-grained patches. Naturally, relation understanding is highly important for learning crossmedia correlation. People focus on not only the alignment between discriminative image regions and key words, but also their relations lying in the visual and textual context. Therefore, in this paper, we propose Multi-level Adaptive Visual-textual Alignment (MAVA) approach with the following contributions. First, we propose cross-media multi-pathway fine-grained network to extract not only the local fine-grained patches as discriminative image regions and key words, but also visual relations between image regions as well as textual relations from the context of sentences, which contain complementary information to exploit fine-grained characteristics within different media types. Second, we propose visual-textual bi-attention mechanism to distinguish the fine-grained information with different saliency from both local and relation levels, which can provide more discriminative hints for correlation learning. Third, we propose cross-media multi-level adaptive alignment to explore global, local and relation alignments. An adaptive alignment strategy is further proposed to enhance the matched pairs of different media types, and discard those misalignments adaptively to learn more precise cross-media correlation. Extensive experiments are conducted to perform image-sentence matching on 2 widely-used cross-media datasets, namely Flickr-30K and MS-COCO, comparing with 10 state-of-the-art methods, which can fully verify the effectiveness of our proposed MAVA approach.

CCL: Cross-modal Correlation Learning With Multigrained Fusion by Hierarchical Network

Cross-modal retrieval has become a highlighted research topic for retrieval across multimedia data such as image and text. A two-stage learning framework is widely adopted by most existing methods based on Deep Neural Network (DNN): The first learning stage is to generate separate representation for each modality, and the second learning stage is to get the cross-modal common representation. However, the existing methods have three limitations: (1) In the first learning stage, they only model intra-modality correlation, but ignore inter-modality correlation with rich complementary context. (2) In the second learning stage, they only adopt shallow networks with single-loss regularization, but ignore the intrinsic relevance of intra-modality and inter-modality correlation. (3) Only original instances are considered while the complementary fine-grained clues provided by their patches are ignored. For addressing the above problems, this paper proposes a cross-modal correlation learning (CCL) approach with multi-grained fusion by hierarchical network, and the contributions are as follows: (1) In the first learning stage, CCL exploits multi-level association with joint optimization to preserve the complementary context from intra-modality and inter-modality correlation simultaneously. (2) In the second learning stage, a multi-task learning strategy is designed to adaptively balance the intra-modality semantic category constraints and inter-modality pairwise similarity constraints. (3) CCL adopts multi-grained modeling, which fuses the coarse-grained instances and fine-grained patches to make cross-modal correlation more precise. Comparing with 13 state-of-the-art methods on 6 widely-used cross-modal datasets, the experimental results show our CCL approach achieves the best performance.

Metamorphic evolution and geochronology of the Dunhuang orogenic belt in the Hongliuxia area, northwestern China

Garnet-bearing mafic granulites and amphibolites from the Hongliuxia area of the southern Dunhuang orogenic belt, northwestern China, commonly occur as lenses or boudinages enclosed within metapelite or marble, which represent the block-in-matrix feature typical of orogenic mélange. Three to four generations of metamorphic mineral assemblages are preserved in these rocks. In the high-pressure amphibolites, prograde mineral assemblages (M1) occur as inclusions (hornblende+plagioclase+quartz±chlorite±epidote±ilmenite) preserved within garnet porphyroblasts, and formed at 550–590°C and 7.7–9.2kbar based on geothermobarometry. The metamorphic peak mineral assemblages (M2) are composed of garnet+hornblende+plagioclase+quartz+clinopyroxene, as well as titanite+zircon+rutile+apatite as accessory minerals in the matrix, and are estimated to have formed at 640–720°C and 14.1–16.0kbar. The first retrograde assemblages (M3) are characterized by “white-eye socket” symplectites (hornblende+plagioclase+quartz±biotite±epidote±magnetite) rimming garnet porphyroblasts, which formed at the expense of the garnet rims and adjacent matrix minerals during the decompression stage under P–T conditions of 610–630°C and 5.6–11.8kbar. The second retrograde assemblages (M4) are intergrowths of actinolite and worm-like quartz produced by the breakdown of the matrix hornblendes, and formed under P–T conditions of ∼490°C and ∼2.8kbar. For the high-pressure mafic granulites, the prograde assemblages (M1) are represented by plagioclase+quartz preserved within the garnet porphyroblasts. The metamorphic peak assemblages (M2) are garnet+matrix minerals (clinopyroxene+plagioclase+quartz+hornblende+rutile+zircon) and were estimated to have formed at ∼680°C and ∼15.4kbar. The retrograde assemblages (M3) are characterized by fine-grained patches of hornblende+plagioclase+quartz rimming the garnet porphyroblasts, as well as hornblende rimming clinopyroxene in the matrix, and were inferred to have formed at ∼620°C and ∼4.2kbar. For the metapelitic gneiss, the metamorphic peak assemblages are the garnet porphyroblasts plus the matrix minerals (biotite+plagioclase+quartz+ilmenite+zircon), which were estimated to have formed at ∼630°C and ∼8.9kbar. The mafic granulites and amphibolites record fairly similar clockwise P–T paths that include nearly isothermal decompression processes, which suggest that they experienced subduction and subsequent rapid tectonic exhumation. SIMS and LA-ICP-MS U–Pb dating of zircons and 40Ar/39Ar dating of hornblende suggest that the metamorphism occurred at ∼430–390Ma. Field occurrences, different protolith ages of the mafic granulites and amphibolites, and the considerable gap in peak P–T conditions between the amphibolite and mafic granulite boudinages and their country rock may suggest a mélange accumulation process during the Paleozoic caused by the Silurian–Devonian orogeny, which is possibly associated with the closure of the Liuyuan ocean, a branch of the Paleo-Asian ocean near the southern Central Asian Orogenic Belt.

Human-Imposed, Fine-Grained Patch Burning Explains the Population Stability of a Fire-Sensitive Conifer in a Frequently Burnt Northern Australia Savanna

Woody plant demographics provide important insight into ecosystem state-shifts in response to changing fire regimes. In Australian tropical savannas, the switch from patchy landscape burning by Aborigines to unmanaged wildfires within the past century has been implicated in biodiversity declines including the fire-sensitive conifer, Callitris intratropica. C. intratropica commonly forms small, closed-canopy groves that exclude fire and allow recruitment of conspecifics and other fire-sensitive woody plants. C. intratropica groves provide a useful indicator of heterogeneity and fire regime change, but the mechanisms driving the species’ persistence and decline remain poorly understood. We examined the hypothesis that C. intratropica population stability depends upon a regime of frequent, low-intensity fires maintained by Aboriginal management. We combined integral projection models of C. intratropica population behaviour with an environmental state change matrix to examine how vital rates, grove dynamics and the frequency of high- and low-intensity fires contribute to population stability. Closed-canopy C. intratropica groves contributed disproportionately to population growth by promoting recruitment, whereas singleton trees accounted for a larger proportion of adult mortality. Our patch-based population model predicted population declines under current fire frequencies and that the recruitment of new groves plays a critical role in the species’ persistence. Our results also indicated that reducing fire intensity, a key outcome of Aboriginal burning, leads to C. intratropica population persistence even at high fire frequencies. These findings provide insight into the relationship between ecosystem composition and human–fire interactions and the role of fire management in sustaining the mosaics that comprise ‘natural’ systems.

Cu-Ni-PGE Mineralization in the South Filson Creek Area, South Kawishiwi Intrusion, Duluth Complex: Mineralization Styles and Magmatic and Hydrothermal Processes

The South Filson Creek mineralization occurs above the basal units of the South Kawishiwi intrusion and represents an uncommon geologic setting of Cu-Ni-platinum-group element (PGE) mineralization within the Duluth Complex. Unlike other mineralization of economic interest in the basal heterogeneous troctolite units along the lower contact, it is located at an approximate stratigraphic height of 1,000 m above the contact with the footwall granitoids in the usually unmineralized homogeneous troctolite unit. Magmatic sulfide mineralization forms disseminated fine-grained patches and pockets in an area of approximately ½ km2 of homogeneous troctolitic rocks in the northeastern part of the South Kawishiwi intrusion. Pyrrhotite, chalcopyrite, pentlandite, and cubanite are the main sulfide minerals with subordinate amounts of other copper-bearing phases (bornite, covellite, talnakhite). Cu/Pd ratios for unaltered mineralization indicate that sulfides were formed from multiple injections of magma that segregated sulfides with different R factors or having different initial compositions. Variation in the halogen contents of apatite from pegmatoids within troctolitic rocks revealed that a fluid has segregated from the crystallizing melt into which fluid Cl and rare earth elements partitioned. Amphibole, chlorite, sericite, prehnite, pumpellyite, and carbonate are associated with alteration along brittle structures and isolated patches in the area. The hydrothermal fluid-rock interaction has affected the composition of primary magmatic sulfide mineralization and is also responsible for mineralization in the anorthositic hanging wall of the South Kawishiwi intrusion. Hydrothermal alteration resulted in a slight Pd depletion in magmatic sulfide mineralization, which might indicate remobilization of precious metals to an unknown location. The hanging-wall mineralization is characterized by relative Cu enrichment in comparison to the magmatic sulfide mineralization with no PGE enrichment. Partial remobilization of platinum group elements (PGE) has also occurred as a result of a separate alteration event, on the evidence that platinum group minerals (PGM) are also associated with serpentinization of olivine and Ca enrichment of plagioclase along grain boundaries with sulfides. This alteration was probably effective on a very local scale.

Spatial relationships and tree species associations across a 236-year boreal mixedwood chronosequence

We studied community development and mortality among tree species at multiple spatial scales in a 236-year mixedwood chronosequence at Lake Duparquet, Quebec. Spatial relationships, species associations, and understory–overstory replacement patterns were studied using spatial statistics, patch indices, G tests, and transition matrices. Results of these analyses showed that shade-intolerant and shade-tolerant tree communities formed increasingly fine-grained patches in midsuccession. These fine-grained associations appear to be partly due to suppressed stems being released into a slowly thinning canopy and partly due to new recruitment of shade-tolerant conifers. In 1-ha plots, tree species richness and evenness peaked in 25-, 100-, and 400-m2 subplots during midsuccession. Forest cover type diversity also peaked in midsuccessional landscapes assessed using aerial photographs. The oldest 1-ha plot was dominated by monospecific patches of eastern white cedar (Thuja occidentalis L.), which was the most likely replacement for live and dead trees and the most abundant species in the numerous subplots that were occupied by trees smaller than 8 cm DBH. In spite of this dominance, long-lived paper birch (Betula papyrifera Marsh.) and residual fir (Abies balsamea (L.) Mill.) seedlings may maintain the mixedwood character of this stand into the future. The implications of our findings for stand management and conservation are discussed.

An Integrated Model of Two Fire Regimes

Casual observers of chaparral in southern California(SCA) quickly learn two key facts about fires: they maybe started in a variety of weather conditions—windy orcalm, dry or humid—and the most spectacular modernburns occur during Santa Ana winds. Serious research-ers who seek to understand ecosystems and guide landmanagers must ask whether this has always been true.Readers should be surprised that Keeley and Fothering-ham (2001 [this issue]) answer yes. First, the chaparralecosystem has certainly not been static: it has experi-enced major changes in fire management as southernCalifornia’s population has increased. Second, observa-tions of the chaparral in nearby Baja California (BCA)immediately reveal a very different dynamic that surelydeserves serious consideration as a model for SCA, onethat precedes the establishment of modern suppressionpractices and dense population. The chaparral of BCA ischaracterized by smaller stands and a propensity forlow-intensity fires in relatively calm, humid weather.The sharp transition between the two regimes cannotbe explained by natural gradients in flora or weather; itfollows the international border, an artificial line drawnby human beings. Vegetation maps show that Californiaecosystems extend 200 km into BCA (Minnich & Franco-Vizcaino 1998).Climatic gradients including temperature and meanannual precipitation cross the border at right angles, andprevailing winds are everywhere westerly. Open-rangecattle grazing is practiced in BCA, but chaparral is unpal-atable to livestock (Minnich & Bahre 1995). Without dis-tinctive suppression systems, changes in fire regime shouldbe expressed in a continuum along environmental gradi-ents, not the discontinuity seen along the border. Theprimary difference is that BCA has not experienced thesame protectionist management policies as SCA, and thedivergence in chaparral fire ecology may provide insightinto the nature of historical vegetation change in SCA.The seminal question is how fires shaped ecosystemswithout management interference. Suppression is un-precedented in ecological history.Keeley and Fotheringham are obviously aware of theneed to balance the BCA and SCA models in any plausi-ble account of chaparral history. Yet their analysis fo-cuses on a 1983 account of the cross-border contrastand ignores a substantial body of more recent publica-tions, including a 52-year transborder fire history (1920–1971; Minnich & Chou 1997), a study of transborder post-fire chronosequences (Minnich & Bahre 1995), and stud-ies of presuppression fire regimes in SCA (Minnich 1987,1988). To evaluate their case for dismissing the BCAmodel, I must review the key findings they ignore. Thesefindings concern the following sequence of topics: (1)the nonrandom turnover of fire patches, (2) the over-abundance of natural ignitions compared with fuel pro-duction; (3) the dominance of landscape burning by rel-atively few fires (most fire starts fail), and (4) therandom phasing of fires with normal weather in the ab-sence of fire control. This is not just an intellectual exer-cise: these are serious implications for fire management.Both the original and subsequent studies (Minnich1983; Minnich & Chou 1997) show that, in the absenceof suppression, the chaparral of BCA is a diverse, fine-grained patch mosaic. With fire suppression in SCA, thechaparral comprises unbroken carpets of mature vegeta-tion interspersed with a few extensively denuded water-sheds (Fig. 1). There are ample data to suggest differ-ences in fire management, including the number of firesof 15 ha (BCA, 2000 events; SCA 350 events), maxi-mum fire size (BCA, 3,000 ha; SCA 59,000 ha), and dif-ferences in fire weather (BCA, onshore flows; SCA, off-shore flows). The only similarity is that fire-return intervalsare 50–70 years in both countries. Our integrated modelexplains the disparate fire outcomes that have existed inthe two countries since the 1920s, Keeley and Fother-ingham do not.An important aspect in developing fire-disturbancetheory is choosing an incontrovertible starting point thatleads in directions productive for research. The modelby Minnich and Chou (1997) is based on the fact that

Wildland Fire Patch Dynamics in the Chaparral of Southern California and Northern Baja California

In ecosystems where fire occurrence has significant time-dependence, fire sequences should exhibit system-regulation that is distinguished by nonrandom (nonstationary), self-organizing patch dynamics related to spatially constrained fire probabilities. Exogenous factors such as fire weather, precipitation variability, and terrain alter the flammability of vegetation and encourage randomness in fire occurrence within pre-existing patch structure. In Californian chaparral, the roles of succession/fuel build-up and exogenous factors is examined by taking advantage of a 100 yr 'natural experiment' in southern California (SCA) and northern Baja California, Mexico (BCA), where factors influencing fire occurrence have been systematically altered by divergent management systems. In SCA, suppression has been practiced since 1900. In BCA, fire control was not official policy until the 1960s and has not been effectively practiced. Fire perimeter histories for 1920-1971 in SCA and BCA, reconstructed from fire history records and repeat aerial photographs, are compared for fire frequency (events/area), size, rotation periods, stand age structure, ignition rates, weather, burning season, and drought. Landscape-scale fire rotation periods are long (≈70 yr) regardless of management policies because fire occurrence is driven by the gradual development of fire hazard during succession, produced by small annual increments of growth and litterfall, as well as by high fuel moisture in evergreen shrubs. Without fire control frequent fires establish fine-grained mosaics. Fire control reduces fire frequencies, increases fire size, and encourages coarse-scale patch structure. Patch dynamics exhibit evidences of nonrandom turnover. Fire size distributions reflect the nearest-neighbor distances between patches below some age-dependent combustion threshold (CT) in the patch mosaic that resist the spread of fires in stands older than CT. Regional burn rates are poorly related to fire frequency, ignition rates, drought, and terrain. The small size of fires in BCA may be reinforced by interactions between fire and pre-existing, fine-grained patch structure, and by random fire occurrence in the probability distributions of fire weather and climate. In SCA, fires are nonrandomly restricted by fire control to extreme weather.

Distribution of proteolipid protein and myelin basic protein in cultured mouse oligodendrocytes: Primary vs. secondary cultures

The distribution of proteolipid protein (PLP) and myelin basic protein (MBP) was examined in differentiating oligodendrocytes of primary and secondary mouse brain cell cultures by single- and double-label indirect immunofluorescence. In primary cultures, MBP and PLP were differentially located in oligodendrocytes. MBP became concentrated as fine punctate dots lining the edges of processes and as coarse grains in flattened sheet-like structures. PLP was distributed diffusely throughout cell bodies and processes but was limited to the perimeter of sheets and some processes within sheets. To compare the detailed distribution of PLP and MBP in the absence of underlying cells, a simple method for the growth of isolated oligodendrocytes in secondary cultures was developed. Cells were maintained in primary culture for 39-41 days, harvested by scraping, enriched for oligodendrocytes, and plated at low cell density. After 1 week, isolated oligodendrocytes had developed long processes and large flattened membranous sheets. MBP and PLP were differentially localized in these cell structures. The sheets contained fine-grained patches of MBP, which were surrounded by networks of MBP- processes. In contrast, PLP was initially seen throughout the cell bodies and processes. In older cultures, PLP became strikingly concentrated in curvilinear membranous profiles. The observations show that PLP and MBP are differentially located in cultured mouse oligodendrocytes. Furthermore, the precise distribution of these myelin-specific antigens is dependent on culture conditions.