2 May 2025 in IEEE Transactions on Visualization and Computer Graphics
Sketch-based modeling studies reconstructing models from sketches automatically, allowing users visualize design concepts rapidly. Generating CAD models based on user sketches helps reduce the learning curve for novice users, which promotes the everyday use of CAD software, and expands its reach to non-professional groups. While various algorithms study automatically generating models from single sketch or line drawing, they often produce non-editable models or editable models limited to simple extrusion operations. To improve this issue, we propose a novel sketch-based modeling system, Sketch2Seq, which generates complex, semantic, and editable CAD models. Our system eliminates the need for additional annotations from users and produces models that support subsequent application in commercial software. The core of our method lies in understanding users' design intent from CAD sketches. We design a novel sketch segmentation network for identifying diverse operation features in CAD sketches, which utilizes geometric features of strokes and different levels of topological connections. Additionally, to tackle the segmentation task, a dataset for CAD sketch segmentation is introduced. Comparative experiments and ablation evaluations prove the effectiveness of the proposed method. Based on segmentation result, coarse CAD sequences are generated and progressively executed. Meanwhile, the orders and parameters of the CAD sequences are optimized with context models and input sketches. All algorithms are integrated into a user interface. Experiments and evaluations validate the feasibility and superiority of our entire system which is able to reconstruct more complex features and achieve better results for longer sequence.
14 August 2024 in IEEE Robotics and Automation Letters
How to estimate camera pose from motion-blurred images remains a challenge for visual odometry. The blurring artifacts are inevitably caused by the exposure during camera motion. While current visual odometry regards them as noise, we argue that it is necessary to extract potential information from blur artifacts, as they contain prior knowledge of camera motion. Base on this, we propose a blur-robust visual odometry that improves the accuracy of camera pose estimation through exposure trajectory. Specifically, we first use the exposure trajectory to guide pixel matching between neighboring frames. The blur mask is then generated based on the magnitude of the exposure trajectory. The mask makes the pose module pay less attention to the feature information in the severely blurred regions. Experiments show that our proposed end-to-end visual odometry achieves competitive performance on most sequences of motion blurred datasets.