While diffusion models have significantly advanced the quality of image generation, their capability to accurately and coherently render text within these images remains a substantial challenge. Conventional diffusion-based methods for scene text generation are typically limited by their reliance on an intermediate layout output. This dependency often results in a constrained diversity of text styles and fonts, an inherent limitation stemming from the deterministic nature of the layout generation phase. To address these challenges, this paper introduces Scene TextGen, a novel diffusion-based model specifically designed to circumvent the need for a predefined layout stage. By doing so, Scene-TextGen facilitates a more natural and varied representation of text. The novelty of SceneTextGen lies in its integration of three key components: a character-level encoder for capturing detailed typographic properties, coupled with a character-level instance segmentation model and a word-level spotting model to address the issues of unwanted text generation and minor character inaccuracies. We validate the performance of our method by demonstrating improved character recognition rates on generated images across different public visual text datasets in comparison to both standard diffusion based methods and text specific methods.

We introduce Lumos, the first end-to-end multimodal question-answering system with text understanding capabilities. At the core of Lumos is a Scene Text Recognition (STR) component that extracts text from first person point-of-view images, the output of which is used to augment input to a Multimodal Large Language Model (MM-LLM). While building Lumos, we encountered numerous challenges related to STR quality, overall latency, and model inference. In this paper, we delve into those challenges, and discuss the system architecture, design choices, and modeling techniques employed to overcome these obstacles. We also provide a comprehensive evaluation for each component, showcasing high quality and efficiency.

Existing autonomous vehicle (AV) navigation algorithms treat lane recognition, obstacle avoidance, local path planning, and lane following as separate functional modules which result in driving behavior that is incompatible with human drivers. It is imperative to design human-compatible navigation algorithms to ensure transportation safety. We develop a new tightly-coupled perception-planning framework that combines all these functionalities to ensure human-compatibility. Using GPS-camera-lidar sensor fusion, we detect actual lane boundaries (ALBs) and propose availability-reasonability-feasibility (ARF) threefold tests to determine if we should generate virtual lane boundaries (VLBs) or follow ALBs. If needed, VLBs are generated using a dynamically adjustable multi-objective optimization framework that considers obstacle avoidance, trajectory smoothness (to satisfy vehicle kinodynamic constraints), trajectory continuity (to avoid sudden movements), GPS following quality (to execute global plan), and lane following or partial direction following (to meeting human expectation). Consequently, vehicle motion is more human compatible than existing approaches. We have implemented our algorithm and tested under open source data with satisfying results.