Image-to-image translation aims to learn the mapping between two visual domains. There are two main challenges for this task: (1) lack of aligned training pairs and (2) multiple possible outputs from a single input image. In this work, we present an approach based on disentangled representation for generating diverse outputs without paired training images. To synthesize diverse outputs, we propose to embed images onto two spaces: a domain-invariant content space capturing shared information across domains and a domain-specific attribute space. Our model takes the encoded content features extracted from a given input and attribute vectors sampled from the attribute space to synthesize diverse outputs at test time. To handle unpaired training data, we introduce a cross-cycle consistency loss based on disentangled representations. Qualitative results show that our model can generate diverse and realistic images on a wide range of tasks without paired training data. For quantitative evaluations, we measure realism with user study and Fréchet inception distance, and measure diversity with the perceptual distance metric, Jensen-Shannon divergence, and number of statistically-different bins.

Latent code z ImageThis bird has feathers that are black and has a red belly Text Figure 1: Mode seeking generative adversarial networks (MSGANs). (Left) Existing conditional generative adversarial networks tend to ignore the input latent code z and generate images of similar modes. (Right) We propose a simple yet effective mode seeking regularization term that can be applied to arbitrary conditional generative adversarial networks in different tasks to alleviate the mode collapse issue and improve the diversity. AbstractMost conditional generation tasks expect diverse outputs given a single conditional context. However, conditional generative adversarial networks (cGANs) often focus on the prior conditional information and ignore the input noise vectors, which contribute to the output variations. Recent attempts to resolve the mode collapse issue for cGANs are usually task-specific and computationally expensive. In this work, we propose a simple yet effective regularization term to address the mode collapse issue for cGANs. The proposed method explicitly maximizes the ratio of the distance between generated images with respect to the corresponding latent codes, thus encouraging the generators to explore more minor modes during training. This mode seeking regularization term is readily applicable to various conditional generation tasks without imposing training overhead or modifying the original network structures. We validate the proposed algorithm on three conditional image synthesis tasks including categorical generation, image-toimage translation, and text-to-image synthesis with different baseline models. Both qualitative and quantitative results demonstrate the effectiveness of the proposed regularization method for improving diversity without loss of quality.

Accumulating evidence indicates that N 2 O emission factors (EFs) vary with nitrogen additions and environmental variations. Yet the impact of the latter was often ignored by previous EF determinations. We developed piecewise statistical models (PMs) to explain how the N 2 O EFs in agricultural soils depend upon various predictors such as climate, soil attributes, and agricultural management. The PMs are derived from a new Bayesian Recursive Regression Tree algorithm. The PMs were applied to the case of EFs from agricultural soils in China, a country where large EF spatial gradients prevail. The results indicate substantial improvements of the PMs compared with other EF determinations. First, PMs are able to reproduce a larger fraction of the variability of observed EFs for upland grain crops (84%, n = 381) and paddy rice (91%, n = 161) as well as the ratio of EFs to nitrogen application rates (73%, n = 96). The superior predictive accuracy of PMs is further confirmed by evaluating their predictions against independent EF measurements (n = 285) from outside China. Results show that the PMs calibrated using Chinese data can explain 75% of the variance. Hence, the PMs could be reliable for upscaling of N 2 O EFs and fluxes for regions that have a phase space of predictors similar to China. Results from the validated models also suggest that climatic factors regulate the heterogeneity of EFs in China, explaining 69% and 85% of their variations for upland grain crops and paddy rice, respectively. The corresponding N 2 O EFs in 2008 are 0.84 ± 0.18% (as N 2 O-N emissions divided by the total N input) for upland grain crops and 0.65 ± 0.14% for paddy rice, the latter being twice as large as the Intergovernmental Panel on Climate Change Tier 1 defaults. Based upon these new estimates of EFs, we infer that only 22% of current arable land could achieve a potential reduction of N 2 O emission of 50%.

We demonstrate an all-optical terahertz modulator based on single-layer graphene on germanium (GOG), which can be driven by a 1.55 μm CW laser with a low-level photodoping power. Both the static and dynamic THz transmission modulation experiments were carried out. A spectrally wide-band modulation of the THz transmission is obtained in a frequency range from 0.25 to 1 THz, and a modulation depth of 94% can be achieved if proper pump power is applied. The modulation speed of the modulator was measured to be ~200 KHz using a 340 GHz carrier. A theoretical model is proposed for the modulator and the calculation results indicate that the enhanced THz modulation is mainly due to the third order nonlinear effect in the optical conductivity of the graphene monolayer.

No abstract