Selected work

Representative Papers

Lifting the Preprocessor with Oxidize: Structure-Preserving C-to-Rust Translation (Technical Report)

Sep 24, 2026

This study addresses the challenge that conventional macro preprocessing expansion during C-to-Rust transpilation leads to the loss of original code abstraction structures and semantic corruption. To overcome this, we propose Oxidize, a system introducing a novel structure-preserving translation mechanism that directly maps C macros to Rust macros to maintain high-level abstractions. Technically, the approach designs a custom intermediate representation (IR) alongside a macro type system, leveraging auxiliary traits and generic conversion macros to achieve precise transpilation. Experimental evaluations across nine real-world C codebases demonstrate that this method delivers high-fidelity transpilation with minimal performance overhead, effectively resolving the longstanding macro handling difficulties inherent in cross-language migration.

0 citationsRead paper

Diffuse the object, keep its label: curating detector training data from a few unlabeled photographs via VLM-built 3D vegetation scenes

Aug 10, 2026

This work addresses the challenges of scarce training data and poor cross-scenario generalization in small object detection under heavy vegetation occlusion. It proposes a novel annotation-free synthetic data generation method that leverages a small set of unlabeled in-the-wild images to reconstruct 3D vegetated scenes guided by a vision-language model (VLM). Realistic annotated images are synthesized by embedding 3D object meshes into these scenes, followed by controllable diffusion-based inpainting to refine object textures and occlusion patterns. A hierarchical mask-locking mechanism is introduced to enhance recall for minority classes while preserving semantic consistency. Evaluated on a humanitarian demining benchmark, detectors trained solely on this synthetic data match or even surpass models trained with substantially more real annotated data from different domains, demonstrating significantly improved unsupervised cross-domain adaptability.

0 citationsRead paper

Recent publications

Latest Papers

Lifting the Preprocessor with Oxidize: Structure-Preserving C-to-Rust Translation (Technical Report)

Sep 24, 2026

This study addresses the challenge that conventional macro preprocessing expansion during C-to-Rust transpilation leads to the loss of original code abstraction structures and semantic corruption. To overcome this, we propose Oxidize, a system introducing a novel structure-preserving translation mechanism that directly maps C macros to Rust macros to maintain high-level abstractions. Technically, the approach designs a custom intermediate representation (IR) alongside a macro type system, leveraging auxiliary traits and generic conversion macros to achieve precise transpilation. Experimental evaluations across nine real-world C codebases demonstrate that this method delivers high-fidelity transpilation with minimal performance overhead, effectively resolving the longstanding macro handling difficulties inherent in cross-language migration.

0 citationsRead paper

Diffuse the object, keep its label: curating detector training data from a few unlabeled photographs via VLM-built 3D vegetation scenes

Aug 10, 2026

This work addresses the challenges of scarce training data and poor cross-scenario generalization in small object detection under heavy vegetation occlusion. It proposes a novel annotation-free synthetic data generation method that leverages a small set of unlabeled in-the-wild images to reconstruct 3D vegetated scenes guided by a vision-language model (VLM). Realistic annotated images are synthesized by embedding 3D object meshes into these scenes, followed by controllable diffusion-based inpainting to refine object textures and occlusion patterns. A hierarchical mask-locking mechanism is introduced to enhance recall for minority classes while preserving semantic consistency. Evaluated on a humanitarian demining benchmark, detectors trained solely on this synthetic data match or even surpass models trained with substantially more real annotated data from different domains, demonstrating significantly improved unsupervised cross-domain adaptability.

0 citationsRead paper