Search

The contrastive vision–language pre-trained model CLIP, driven by large-scale open-vocabulary image–text pairs, has recently demonstrated remarkable zero-shot generalization capabilities in diverse downstream image tasks, which has made numerous models dominated by the "image pre-training followed by fine-tuning" paradigm exhibit promising results on standard video benchmarks. However, as models scale up, full fine-tuning adaptive strategy for specific tasks becomes difficult in terms of training and storage. In this work, we propose a novel method that adapts CLIP to the video domain for efficient recognition without destroying the original pre-trained parameters. Specifically, we introduce temporal prompts to realize the object of reasoning about the dynamic content of videos for pre-trained models that lack temporal cues. Then, by replacing the direct learning style of prompt vectors with a lightweight reparameterization encoder, the model can be adapted to domain-specific adjustment to learn more generalizable representations. Furthermore, we predefine a Chinese label dictionary to enhance video representation by co-supervision of Chinese and English semantics. Extensive experiments on video action recognition benchmarks show that our method achieves competitive or even better performance than most existing methods with fewer trainable parameters in both general and few-shot recognition scenarios. [ABSTRACT FROM AUTHOR]

The natural gas resources in the Sichuan Basin are abundant and exhibit various phenomena, for example, mixed sources, mixed maturities and overlapping combinations, resulting in great difficulties in identifying natural gas sources. In this study, several groups of thermal simulation experiments were conducted on various types of kerogens using closed and/or semi‐open systems to investigate the characteristics of the carbon isotope compositions of light hydrocarbon gases derived from different typical kerogens. The results showed that the carbon isotope compositions and variation trends of methane and ethane are controlled by the kerogen type, thermal simulation system and thermal evolution degree. The carbon isotope compositions of light hydrocarbon gases derived from different types of kerogens exhibit some differences with increased thermal evolution degree. When the thermal simulation temperature increases from 320 to 550°C, the methane carbon isotope of typically marine source rocks, mainly type I and type II1 kerogens, is slightly higher than or equivalent to that of ethane. However, the ethane carbon isotope of terrestrial source rocks, primarily types II2 and III kerogens, is significantly heavier than methane. By comparison, the evolution trend of carbon isotope compositions of actual natural gas is well consistent with that of light hydrocarbon gases cracked from thermal simulation experiments with increased thermal evolution degree. These findings are of great importance to identify the source rock of actual natural gas with high thermal maturities, as well as to determine the sources of deep shale gas in the southern Sichuan Basin, the Yuanba gas field and the Puguang gas field. Finally, a fitting method of carbon isotope values of methane and ethane of light hydrocarbon gases can be used to more finely differentiate the sources of natural gas in complex areas with mixed sources, mixed maturities and overlapping combinations. [ABSTRACT FROM AUTHOR]

In this paper, we present an efficient method for constructing symmetric subdivision schemes reproducing high-degree polynomials, without solving a system of linear equations. Original symmetric subdivision and its deduced subdivisions have similar increasing characteristics to the family of pseudo-splines from the polynomial reproduction point of view. Several examples are given to illustrate the efficiency of the method. [ABSTRACT FROM AUTHOR]

In recent years, the typical crude oil from the Cambrian Lower Ordovician source rocks has been found in the Tarim Basin, and it is characterized by 13C enrichment. The stable carbon isotope of the oil is concentrated at approximately −28‰, which is 3‰–6‰ heavier than that of the crude oil from the Middle and Upper Ordovician source rocks. The stable carbon isotope between crude oil and its kerogen is characterized by inversion, with an amplitude of approximately 3‰–4‰. However, the origin of this phenomenon remains controversial, which restricts the next exploration of crude oil in the Cambrian–Lower Ordovician in the Tarim Basin. Samples were collected from seven outcrop profiles in the Kuruketage region. The basic geochemical features of the Cambrian strata, including total organic carbon abundance, δ13Corg, δ13Ccarb, and δ18Ocarb, were studied and compared with those of well TD2. The results show that the lithologies, isotope stratigraphic evolutional curves, and correlations between the organic carbon and inorganic carbon of the profiles from the northern Kuruketage region are successively comparable with those of the profiles from the southern Kuruketage region and the well TD2. In this study, the carbon isotope records are supposed to indicate a vertical gradient Cambrian oceanic mode, including the shallow decoupled isotope zone, middle gradient‐coupled isotope zone, and deep isotope‐decoupled zone. The discovery of this isotope record reflects the formation of stable carbon and oxygen isotope stratified oceans during the Cambrian period. Additionally, it provides important insights into the origin of the 13C enrichment observed in some Lower Cambrian crude oils in the Tarim Basin. [ABSTRACT FROM AUTHOR]

The occurrence of coal-bearing strata in China is characterized by the variety of coal-bearing basins，the remarkable spatial and temporal differences of late tectonic deformations.The main controlling factors，which determine the tectonic framework of coalfields，include geodynamic environment，tectonic evolution，deep structures，tectonic stress，and lithologic combination features of coal measures.The continent of China has experienced multi-stage tectonic movements since the late Paleozoic，and due to the spatio-temporal heterogeneity of continental tectonic evolution and the complexity of basement properties，the coalfield structures present complex and ordered patterns.The coalfield tectonic framework of China can be divided into two major coalfield regions，three coal deformation zones，and five coal occurrence structural areas.Taking coal occurrence structural sub-areas as basic units，the tectonic characteristics of five coal occurrence structural areas in China were summarized in this paper.

The local descriptors based on a binary pattern feature have state-of-the-art distinctiveness. However, their high dimensionality resists them from matching faster and being used in a low-end device. In this paper we propose an efficient and feasible learning method to select discriminative binary patterns for constructing a compact local descriptor. In the selection, a searching tree with Branch&Bound is used instead of the exhaustive enumeration, in order to avoid tremendous computation in training. New local descriptors are constructed based on the selected patterns. The efficiency of selecting binary patterns has been confirmed by the evaluation of these new local descriptors’ performance in experiments of image matching and object recognition.

Multi-target tracking is a challenge due to the variable number of targets and the frequent interaction between targets in complex dynamic environments. This paper presents a multi-target tracking algorithm based on bipartite graph matching. Unlike previous approaches, the method proposed considers the target tracking as a bipartite graph matching problem where the nodes of the bipartite graph correspond to the targets in two neighboring frames, and the edges correspond to the degree of the similarity measure between the targets in different frames. Finding correspondence between the targets is formulated as a maximal matching problem which can be solved by the dynamic Hungarian algorithm. Then, merging and splitting of the targets detection is proposed, the candidate occlusion region is predicted according to the overlapping between the bounding boxes of the interacting targets to handle the mutual occlusion problem. The extensive experimental results show that the algorithm proposed can achieve good performance on dynamic target interactions compared to state-of-the-art methods.