Search

Rain-by-snow weather removal is a specialized task in weather-degraded image restoration aiming to eliminate coexisting rain streaks and snow particles. In this paper, we propose RSFormer, an efficient and effective Transformer that addresses this challenge. Initially, we explore the proximity of convolution networks (ConvNets) and vision Transformers (ViTs) in hierarchical architectures and experimentally find they perform approximately at intra-stage feature learning. On this basis, we utilize a Transformer-like convolution block (TCB) that replaces the computationally expensive self-attention while preserving attention characteristics for adapting to input content. We also demonstrate that cross-stage progression is critical for performance improvement, and propose a global-local self-attention sampling mechanism (GLASM) that down-/up-samples features while capturing both global and local dependencies. Finally, we synthesize two novel rain-by-snow datasets, RSCityScape and RS100K, to evaluate our proposed RSFormer. Extensive experiments verify that RSFormer achieves the best trade-off between performance and time-consumption compared to other restoration methods. For instance, it outperforms Restormer with a 1.53% reduction in the number of parameters and a 15.6% reduction in inference time. Datasets, source code and pre-trained models are available at \url{https://github.com/chdwyb/RSFormer}., code is available at \url{https://github.com/chdwyb/RSFormer}

The advents of Large Language Models (LLMs) have shown promise in augmenting programming using natural interactions. However, while LLMs are proficient in compiling common usage patterns into a programming language, e.g., Python, it remains a challenge how to edit and debug an LLM-generated program. We introduce ANPL, a programming system that allows users to decompose user-specific tasks. In an ANPL program, a user can directly manipulate sketch, which specifies the data flow of the generated program. The user annotates the modules, or hole with natural language descriptions offloading the expensive task of generating functionalities to the LLM. Given an ANPL program, the ANPL compiler generates a cohesive Python program that implements the functionalities in hole, while respecting the dataflows specified in sketch. We deploy ANPL on the Abstraction and Reasoning Corpus (ARC), a set of unique tasks that are challenging for state-of-the-art AI systems, showing it outperforms baseline programming systems that (a) without the ability to decompose tasks interactively and (b) without the guarantee that the modules can be correctly composed together. We obtain a dataset consisting of 300/400 ARC tasks that were successfully decomposed and grounded in Python, providing valuable insights into how humans decompose programmatic tasks. See the dataset at https://iprc-dip.github.io/DARC.

The dense rain accumulation in heavy rain can significantly wash out images and thus destroy the background details of images. Although existing deep rain removal models lead to improved performance for heavy rain removal, we find that most of them ignore the detail reconstruction accuracy of rain-free images. In this paper, we propose a dual-stage progressive enhancement network (DPENet) to achieve effective deraining with structure-accurate rain-free images. Two main modules are included in our framework, namely a rain streaks removal network (R$^2$Net) and a detail reconstruction network (DRNet). The former aims to achieve accurate rain removal, and the latter is designed to recover the details of rain-free images. We introduce two main strategies within our networks to achieve trade-off between the effectiveness of deraining and the detail restoration of rain-free images. Firstly, a dilated dense residual block (DDRB) within the rain streaks removal network is presented to aggregate high/low level features of heavy rain. Secondly, an enhanced residual pixel-wise attention block (ERPAB) within the detail reconstruction network is designed for context information aggregation. We also propose a comprehensive loss function to highlight the marginal and regional accuracy of rain-free images. Extensive experiments on benchmark public datasets show both efficiency and effectiveness of the proposed method in achieving structure-preserving rain-free images for heavy rain removal. The source code and pre-trained models can be found at \url{https://github.com/wybchd/DPENet}.

Calligraphy plays a very important role in the history of China. From ancient times to modern times, the beauty of calligraphy has been passed down to the present. Different calligraphy styles and structures have made calligraphy a beauty and embodiment in the field of writing. However, the recognition of calligraphy style and fonts has always been a blank in the computer field. The structural complexity of different calligraphy also brings a lot of challenges to the recognition technology of computers. In my research, I mainly discussed some of the main recognition techniques and some popular machine learning algorithms in this field for more than 20 years, trying to find a new method of Chinese calligraphy styles recognition and exploring its feasibility. In our research, we searched for research papers 20 years ago. Most of the results are about the content recognition of modern Chinese characters. At first, we analyze the development of Chinese characters and the basic Chinese character theory. In the analysis of the current recognition of Chinese characters (including handwriting online and offline) in the computer field, it is more important to analyze various algorithms and results, and to analyze how to use the experimental data, besides how they construct the data set used for their test. The research on the method of image processing based on Chinese calligraphy works is very limited, and the data collection for calligraphy test is very limited also. The test of dataset that used between different recognition technologies is also very different. However, it has far-reaching significance for inheriting and carrying forward the traditional Chinese culture. It is very necessary to develop and promote the recognition of Chinese characters by means of computer tecnchque. In the current application field, the font recognition of Chinese calligraphy can effectively help the library administrators to identify the problem of the classification of the copybook, thus avoiding the recognition of the calligraphy font which is difficult to perform manually only through subjective experience. In the past 10 years of technology, some techniques for the recognition of single Chinese calligraphy fonts have been given. Most of them are the pre-processing of calligraphy characters, the extraction of stroke primitives, the extraction of style features, and the final classification of machine learning. The probability of the classification of the calligraphy works. Such technical requirements are very large for complex Chinese characters, the result of splitting and recognition is very large, and it is difficult to accurately divide many complex font results. As a result, the recognition rate is low, or the accuracy of recognition of a specific word is high, but the overall font recognition accuracy is low. We understand that Chinese calligraphy is a certain research value. In the field of recognition, many research papers on the analysis of Chinese calligraphy are based on the study of calligraphy and stroke. However, we have proposed a new method for dealing with font recognition. The recognition technology is based on the whole page of the document. It is studied in three steps: the first step is to use Fourier transform and some Chinese calligraphy images and analyze the results. The second is that CNN is based on different data sets to get some results. Finally, we made some improvements to the CNN structure. The experimental results of the thesis show that the full-page documents recognition method proposed can achieve high accuracy with the support of CNN technology, and can effectively identify the different styles of Chinese calligraphy in 5 styles. Compared with the traditional analysis methods, our experimental results show that the method based on the full-page document is feasible, avoiding the cumbersome font segmentation problem. This is more efficient and more accurate.