A total of 219 agricultural soil and 48 vegetable samples were collected from the midstream and downstream of the Xiangjiang River (the Hengyang-Changsha section) in Hunan Province. The accumulation characteristics, spatial distribution and potential risk of heavy metals in the agricultural soils and vegetables were depicted. There are higher accumulations of heavy metals such as As, Cd, Cu, Ni, Pb and Zn in agricultural soils, and the contents of Cd (2.44 mg kg −1 ), Pb (65.00 mg kg −1 ) and Zn (144.13 mg kg −1 ) are 7.97, 3.69 and 1.63 times the corresponding background contents in soils of Hunan Province, respectively. 13.2% of As, 68.5% of Cd, 2.7% of Cu, 2.7% of Ni, 8.7% of Pb and 15.1% of Zn in soil samples from the investigated sites exceeded the maximum allowable heavy metal contents in the China Environmental Quality Standard for Soils (GB15618-1995, Grade II). The pollution characteristics of multi-metals in soils are mainly due to Cd. The contents of As, Cd, Cu, Pb and Zn in vegetable soils are significantly higher than the contents in paddy soils. 95.8%, 68.8%, 10.4% and 95.8% of vegetable samples exceeded the Maximum Levels of Contaminants in Foods (GB2762-2005) for As, Cd, Ni and Pb concentrations, respectively. There are significantly positive correlations between the concentrations of Cd, Pb and Zn in vegetables and the concentrations in the corresponding vegetable soils (p<0.01). It is very necessary to focus on the potential risk of heavy metals for food safety and human health in agricultural soils and vegetables in the midstream and downstream of the Xiangjiang River, Hunan Province of China.

Model inversion, whose goal is to recover training data from a pre-trained model, has been recently proved feasible. However, existing inversion methods usually suffer from the mode collapse problem, where the synthesized instances are highly similar to each other and thus show limited effectiveness for downstream tasks, such as knowledge distillation. In this paper, we propose Contrastive Model Inversion (CMI), where the data diversity is explicitly modeled as an optimizable objective, to alleviate the mode collapse issue. Our main observation is that, under the constraint of the same amount of data, higher data diversity usually indicates stronger instance discrimination. To this end, we introduce in CMI a contrastive learning objective that encourages the synthesizing instances to be distinguishable from the already synthesized ones in previous batches. Experiments of pre-trained models on CIFAR-10, CIFAR-100, and Tiny-ImageNet demonstrate that CMI not only generates more visually plausible instances than the state of the arts, but also achieves significantly superior performance when the generated data are used for knowledge distillation. Code is available at https://github.com/zju-vipa/DataFree.

No abstract

No abstract

No abstract