Deep learning is revolutionizing the mapping industry. Under lightweight human curation, computer has generated almost half of the roads in Thailand on Open-StreetMap (OSM) using high resolution aerial imagery. Bing maps are displaying 125 million computer generated building polygons in the U.S. While tremendously more efficient than manual mapping, one cannot map out everything from the air. Especially for roads, a small prediction gap by image occlusion renders the entire road useless for routing. Misconnections can be more dangerous. Therefore computer based mapping often requires local verifications, which is still labor intensive. In this paper, we propose to leverage crowd sourced GPS data to improve and support road extraction from aerial imagery. Through novel data augmentation, GPS rendering, and 1D transpose convolution techniques, we show almost 5% improvements over previous competition winning models, and much better robustness when predicting new areas without any new training data or domain adaptation.

Deep learning has achieved promising results in pavement distress detection. However, the training model's effectiveness varies according to the data and scenarios acquired by different camera types and their installation positions. It is time consuming and labor intensive to recollect labeled data and retrain a new model every time the scene changes. In this paper, we propose a transfer learning pipeline to address this problem, which enables a distress detection model to be applied to other untrained scenarios. The framework consists of two main components: data transfer and model transfer. The former trains a generative adversarial network to transfer existing image data into a new scene style. Then, attentive CutMix and image melding are applied to insert distress annotations to synthesize the new scene's labeled data. After data expansion, the latter step transfers the feature extracted by the existing model to the detection application of the new scene through domain adaptation. The effects of varying degrees of knowledge transfer are also discussed. The proposed method is evaluated on two data sets from two different scenes with more than 40,000 images totally. This method can reduce the demand for training data by at least 25% when the model is applied in a new scene. With the same number of training images, the proposed method can improve the model accuracy by 26.55%.

Third-party libraries are crucial to the development of software projects. To get suitable libraries, developers need to search through millions of libraries by filtering, evaluating, and comparing. The vast number of libraries places a barrier for programmers to locate appropriate ones. To help developers, researchers have proposed automated approaches to recommend libraries based on library usage pattern. However, these prior studies can not sufficiently match user requirements and suffer from cold-start problem. In this work, we would like to make recommendations based on requirement descriptions to avoid these problems. To this end, we propose a novel neural approach called Req2Lib which recommends libraries given descriptions of the project requirement. We use a Sequence-to-Sequence model to learn the library linked-usage information and semantic information of requirement descriptions in natural language. Besides, we apply a domain-specific pre-trained word2vec model for word embedding, which is trained over textual corpus from Stack Overflow posts. In the experiment, we train and evaluate the model with data from 5,625 java projects. Our preliminary evaluation demonstrates that Req2Lib can recommend libraries accurately.

No abstract