We present an efficient 3D object detection framework based on a single RGB image in the scenario of autonomous driving. Our efforts are put on extracting the underlying 3D information in a 2D image ...and determining the accurate 3D bounding box of object without point cloud or stereo data. Leveraging the off-the-shelf 2D object detector, we propose an artful approach to efficiently obtain a coarse cuboid for each predicted 2D box. The coarse cuboid has enough accuracy to guide us to determine the 3D box of the object by refinement. In contrast to previous state-of-the-art methods that only use the features extracted from the 2D bounding box for box refinement, we explore the 3D structure information of the object by employing the visual features of visible surfaces. The new features from surfaces are utilized to eliminate the problem of representation ambiguity brought by only using 2D bounding box. Moreover, we investigate different methods of 3D box refinement and discover that a classification formulation with quality aware loss have much better performance than regression. Evaluated on KITTI benchmark, our approach outperforms current state-of-the-art methods for single RGB image based 3D object detection.
Cascaded classifiers have been widely used in pedestrian detection and achieved great success. These classifiers are trained sequentially without joint optimization. In this paper, we propose a new ...deep model that can jointly train multi-stage classifiers through several stages of back propagation. It keeps the score map output by a classifier within a local region and uses it as contextual information to support the decision at the next stage. Through a specific design of the training strategy, this deep architecture is able to simulate the cascaded classifiers by mining hard samples to train the network stage-by-stage. Each classifier handles samples at a different difficulty level. Unsupervised pre-training and specifically designed stage-wise supervised training are used to regularize the optimization problem. Both theoretical analysis and experimental results show that the training strategy helps to avoid over fitting. Experimental results on three datasets (Caltech, ETH and TUD-Brussels) show that our approach outperforms the state-of-the-art approaches.
Part-based models have demonstrated their merit in object detection. However, there is a key issue to be solved on how to integrate the inaccurate scores of part detectors when there are occlusions, ...abnormal deformations, appearances, or illuminations. To handle the imperfection of part detectors, this paper presents a probabilistic pedestrian detection framework. In this framework, a deformable part-based model is used to obtain the scores of part detectors and the visibilities of parts are modeled as hidden variables. Once the occluded parts are identified, their effects are properly removed from the final detection score. Unlike previous occlusion handling approaches that assumed independence among the visibility probabilities of parts or manually defined rules for the visibility relationship, a deep model is proposed in this paper for learning the visibility relationship among overlapping parts at multiple layers. The proposed approach can be viewed as a general postprocessing of part-detection results and can take detection scores of existing part-based models as input. The experimental results on three public datasets (Caltech, ETH, and Daimler) and a new CUHK occlusion dataset (<;uri xlink:type="simple">http://www.ee.cuhk.edu.hk/~xgwang/CUHK_pedestrian.html<;/uri>), which is specially designed for the evaluation of occlusion handling approaches, show the effectiveness of the proposed approach.
CdIn
2
S
4
and zinc tetrakis(4-carboxyphenyl)porphyrin (ZnTCPP) were synthesized by hydrothermal method, and an organic dye-sensitized inorganic semiconductor ZnTCPP/CdIn
2
S
4
type II heterojunction ...was constructed on a fluorine-doped tin oxide (FTO) substrate electrode. A sandwich immunostructure for signal-attenuation photoelectrochemical (PEC) detection of cardiac troponin I (cTnI) was constructed using the ZnTCPP/CdIn
2
S
4
/FTO photoanode and a horseradish peroxidase (HRP)-ZnFe
2
O
4
-Ab
2
-bovine serum albumin (BSA) immunolabeling complex. The bioenzyme HRP and the HRP-like nanozyme ZnFe
2
O
4
can co-catalyze the oxidation of 4-chloro-1-naphthol (4-CN) by H
2
O
2
to produce an insoluble precipitate on the photoanode, thus notably reducing the anodic photocurrent for quantitative determination of cTnI. Under the optimal conditions, the photocurrent at 0 V vs. SCE in 0.1 M phosphate buffer solution (pH 7.40) containing 0.1 M ascorbic acid was linear with the logarithm of cTnI concentration from 500 fg mL
−1
to 50.0 ng mL
−1
, and the limit of detection (LOD,
S
/
N
= 3) is 0.15 pg mL
−1
. Spiked recoveries were 95.1% ~ 104% for assay of cTnI in human serum samples.
Graphical Abstract
•A prediction framework for plain wetland biodiversity is proposed.•The framework is based on hydrological connectivity and land use prediction.•Strategies for sustainable wetland management were put ...forward.
Wetland destruction and degradation have been increasing gradually. The biodiversity of the remaining wetlands is under unprecedented threat. Identifying the future trend of wetland biodiversity is a critical step to provide early warnings for wetland biodiversity protection and management. However, studies have focused mainly on the current biodiversity assessment and protection, without emphasizing on the prediction of future changes. Combining the advantages of the key indicators of wetland biodiversity simulation (wetland pattern and hydrological connectivity; PHC) and CA_Markov of land use prediction, this study proposes a prediction framework for wetland biodiversity. Taking Sanjiang Plain as an example, this study predicted the changing trend of wetland biodiversity in the study area and evaluated the potential loss of wetland biodiversity in each reserve. The results revealed that the cultivated land occupation mainly caused the change in the spatial patterns of wetland biodiversity in the study area. According to the land use development trend from 2010 to 2015, the indexes of wetland PHC in the study area will decline significantly from 2020 to 2030, and the wetland biodiversity predicted by our framework will be transformed from the medium to the low level (the biodiversity conservation value will decrease 7.40% on average, with the wetland area reduced by 2.74%). Each reserve in the study area will experience various degrees of degradation in biodiversity due to the decrease in hydrological connectivity. The framework of wetland biodiversity prediction proposed in this study can provide technical support for predicting the changing trend in wetland biodiversity at the regional scale and a reference for long-term protection and monitoring strategies of wetland biodiversity at the reserve scale as an early warning.
In this paper, we tackle the problem of no-reference image quality assessment (IQA). A learning-based IQA framework "VIDGIQA" is proposed, which extracts quality features from the input image and ...regresses the visual quality on these features. Since different distortions lead to different visual perceptions in the human visual system, distortion information is adopted to guide the feature learning process together with the human quality scores. Besides, a regression method is proposed to model and estimate the visual importance weights of all local regions, which can effectively improve the performance. More importantly, all these operations are integrated into one deep neural network, so that they can be jointly optimized and well cooperate with each other. Experiments were conducted to demonstrate the power of the proposed method on several datasets, including the LIVE dataset 1, the TID 2013 dataset 2, the LIVE multiply distorted IQA dataset 3, CSIQ 4 , and the LIVE in the wild image quality database 5. The proposed method achieves 0.969 and 0.973 on the LIVE dataset 1 in terms of the spearman rank-order correlation coefficient and the Pearson linear correlation coefficient, respectively, which outperforms the state-of-the-art methods.
In this paper, we propose deformable deep convolutional neural networks for generic object detection. This new deep learning object detection framework has innovations in multiple aspects. In the ...proposed new deep architecture, a new deformation constrained pooling (def-pooling) layer models the deformation of object parts with geometric constraint and penalty. A new pre-training strategy is proposed to learn feature representations more suitable for the object detection task and with good generalization capability. By changing the net structures, training strategies, adding and removing some key components in the detection pipeline, a set of models with large diversity are obtained, which significantly improves the effectiveness of model averaging. The proposed approach improves the mean averaged precision obtained by RCNN 14, which was the state-of-the-art, from 31% to 50.3% on the ILSVRC2014 detection test set. It also outperforms the winner of ILSVRC2014, GoogLeNet, by 6.1%. Detailed component-wise analysis is also provided through extensive experimental evaluation, which provide a global view for people to understand the deep learning object detection pipeline.
•The landscape pattern of farming-land has impacts on soil- and vegetation- indicators of wetland.•The degradation of remnant wetland (IDRW) can be modeled by soil- and vegetation- indicators.•The ...landscape pattern of surrounding cultivated-land can present IDRW by simulation.
Human activities, especially cultivated-land reclamation, substantially influence landscape patterns. This phenomenon leads to a sharp decrease in wetland areas and wetland function degradation. To understand the impact of changing landscape patterns of the cultivated land on remnant wetland degradation and to formulate highly targeted control measures, considering a farmland-wetland mosaic landscape in the Sanjiang Plain as an example, this study explored the response of soil- and vegetation-related indicators of remnant wetlands under the influence of changing cultivated land patterns. Correlations between cultivated land patterns and soil- and vegetation-indicators in different buffer zones of remnant wetland sample points were discussed. On the basis of modeling an index of the integrated degradation degree of remnant wetlands (IDRWs) using soil- and vegetation-indicators, the correlation between cultivated land patterns and IDRW was analysed. The results showed that: 1) five soil-related indexes of remnant wetlands were lower than those of natural wetlands, and nine vegetation-related indexes of remnant wetlands were higher than those of natural wetlands; 2) the variation in soil- and vegetarian-related indexes caused by changing cultivated land patterns in a certain range of remnant wetlands was significant (P < 0.05) and these indexes could potentially be used as variables to model IDRW; and 3) the aggregation degree of wetland patches was the main landscape index for IDRW simulation (from 1995 to 2015, the proportion of serious degradation of the Sanjiang Plain wetland first decreased in 2010 and then increased, and that of slight degradation first increased in 2010 and then decreased, indicating that wetland protection and restoration plans have shown progress after 2010). Our study discussed the impacts of cultivated land patterns on remnant wetland degradation and developed a simulation method of degraded remnant wetlands to provide a reference for the wetland protection, restoration, degradation monitoring, and reclamation planning for the cultivated lands in the future.
Large-scale human activities especially the destruction of forest land, grassland, and unused land result in a large amount of carbon release into the atmosphere and cause drastic changes in land ...use/cover in the Sanjiang Plain. As a climate change-sensitive and ecologically vulnerable area, the Sanjiang Plain ecosystem’s carbon cycle is affected by significant climate change. Therefore, it is important that studying the impact of the changes in land use/cover and climate on vegetation carbon storage in the Sanjiang Plain. Remote sensing, temperature, and precipitation data in four periods from 2001 to 2015 are used as bases in conducting an analysis of land use/cove types and spatio-temporal variation of vegetation carbon density and carbon storage in growing season using model and related analysis methods. Moreover, the impact of land use/cover change and climate change on vegetation carbon density and carbon storage is discussed. The findings are as follows. (1) Cultivated land in the Sanjiang Plain increased, while forest land, grassland and unused land generally decreased. (2) Vegetation carbon density increased, in which the average carbon density of cultivated land, grassland, and unused land varied insignificantly, while that of forest land increased continuously from 4.18 kg C/m
2
in 2001 to 7.65 kg C/m
2
in 2015. Vegetation carbon storage increased from 159.18 Tg C in 2001 to 256.83 Tg C in 2015, of which vegetation carbon storage of forest land contributed 94% and 97%, respectively. (3) Conversion of land use/cover types resulted in a 22.76-TgC loss of vegetation carbon storage. Although the forest land area decreased by 3389.5 km
2
, vegetation carbon storage in the research area increased by 97.65 Tg C owing to the increase of forest carbon density. (4) Pixel-by-pixel analysis showed that vegetation carbon storage in the majority of the areas of the Sanjiang Plain are negatively correlated with temperature and positively correlated with precipitation. The results showed that changes of land use/cover types and vegetation carbon density directly lead to a change in vegetation carbon storage, with the change of forest vegetation carbon density being the main driver affecting vegetation carbon storage variation. The increase of temperature mainly suppresses the vegetation carbon density, and the increase of precipitation mainly promotes it.