Search

In recent years, generative adversarial networks has made remarkable progress in the field of text-to-image synthesis whose task is to obtain high-quality generated images. Current evaluation metrics in this field mainly evaluate the quality distribution of the generated image dataset rather than the quality of single image itself. With the deepening research of text-to-image synthesis, the quality and quantity of generated images will be greatly improved. There will be a higher demand for generated image evaluation. Therefore, this paper proposes a blind generated image evaluator(BGIE) based on BRISQUE model and sparse neighborhood co-occurrence matrix, which is specially used to evaluate the quality of single generated image. Through experiments, BGIE surpasses all no-reference methods proposed in the past. Compared to VSS method, the surpassing ratio: SRCC is 8.8%, PLCC is 8.8%. By the “One-to-Multi” high-score image screening experiment, it is proved that the BGIE model can screen out best image from multiple images.

The image quality assessment (IQA) for generated images main focuses on the quality of perceptual aspects. Existing methods for evaluating generated images consider the overall image distribution characteristics. At present, there is no practical method for a single generated image. To address this issue, this paper proposes a solution base on the deep feature compatibility (DFC), which first collects suitable comparison images by a collection model. Then it provides an individual score by computing the compatibility of target and pictures with good perceptual quality. This method makes up for the deficiency of Inception Score (IS) in a small number of results or/and a single image. The experiment on Caltech UCSD birds 200 (CUB) shows that our method performs well on the assessment mission for generated images. Finally, we analyze the various problems of the representative IQA methods in evaluating.

α-Linolenic acid (ALA) is an essential fatty acid which cannot be synthesized de novo in mammals and must be ingested regularly in the diet. In this study, a microalgal strain named HDMA-11 was isolated from Lake Ming, China, and was found to accumulate a high ALA content (39.2% of total lipids). Phylogenetic neighbor-joining analysis indicated that HDMA-11 belongs to the genus Monoraphidium (Selenastraceae, Sphaeropleales) and its 18S ribosomal DNA sequence seemed to be a new molecular record of a Monoraphidium species. The fatty acid profiles, biomass productivity and lipid content of HDMA-11 were also investigated in autotrophic conditions. The high levels of polyunsaturated fatty acids in HDMA-11, especially ALA, make it suitable as a source of nutritional supplementation for human health. Furthermore, HDMA-11 exhibited good properties for biodiesel production, characterized by high lipid content (28.5% of dry weight), moderate biomass productivity (31.5 mg L−1 day−1) and a promising lipid profile.

We propose a novel and simple text-to-image synthesizer (MD-GAN) using multiple discrimination. Based on the Generative Adversarial Network (GAN), we introduce segmentation images to the discriminator to ensure the improvement of discrimination ability. The improvement of discrimination ability will enhance the generator’s generating ability, thus obtaining high-resolution results. Experiments well validate the outstanding performance of our algorithm. On CUB dataset, our inception score is 27.7% and 1.7% higher than GAN-CLS-INT and GAWWN, respectively. On the flower dataset, it further outplays GAN-CLS-INT and StackGAN by 21.8% and 1.25%, respectively. At the same time, our model is more concise in structure, and its training time is only half that of StackGAN.

Effective analysis of remote sensing images remains a challenging topic in deep learning research field. This paper proposes a semantic segmentation approach based on the inception architecture. Specifically, the combination of residual network and Inception improves the feature extraction capability of encoder network. In addition, this paper discusses the problem of unbalanced information allocation caused by concatenated operation in the up-sampling process of network models such as Unet, and the RCSE module is proposed to complete feature reconstruction to solve this problem. The approach ensures accurate assignment of semantic labels to the buildings in the aerial images. Experiments based on the dataset proposed by CrowdAi certify the effectiveness of our approach with a 3.7% IoU improvement compared to Unet.

Image captioning is a challenging problem in computer vision, which has numerous practical applications. Recently, the method of dense image captioning has emerged, which realizes the full understanding of the image by localizing and describing multiple salient regions covering the image. Despite there are state-of-the-art approaches encouraging progress, the ability to position and to describe the target area correspondingly is not enough as we expect. To alleviate this challenge, a precise feature extraction method (PFE) is proposed in this paper to further enhance the effect of dense image captioning. Our model is evaluated on the Visual Genome dataset. It demonstrated that our method is better than other state-of-the-art methods.

In this paper, the method of two-stage generation and two-stage discrimination (2G2D) is proposed to generate high-resolution and more realistic images. It is a simple but effective way to synthesize images based on text descriptions. Our method generates the refined foreground image in the first stage, and then combines the text description to generate the final high-resolution image in second stage. We demonstrate the performance of the proposed method on the Caltech-UCSD Birds (CUB) dataset. Through the experimental results, our model can improve the resolution and the authenticity of content of the synthetic image better than the existing state-of-the-art methods.