{"paper":{"title":"Generative Texture Diversification of 3D Pedestrians for Robust Autonomous Driving Perception","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Synthesizing diverse facial textures on one 3D pedestrian base asset improves 2D detection robustness but exposes geometric sensitivities in 3D point-cloud models.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Ahmed Abdullah, Arka Bhowmick, Enes Ozeren, Oliver Wasenmuller","submitted_at":"2026-05-13T16:35:50Z","abstract_excerpt":"In recent years, autonomous driving has significantly in creased the demand for high-quality data to train 2D and 3D perception models for safety-critical scenarios. Real world datasets struggle to meet this demand as require ments continuously evolve and large-scale annotated data collection remains costly and time-consuming making syn thetic data a scalable, practical and controllable alterna tive. Pedestrian detection is among the most safety-critical tasks in autonomous driving. In this paper, we propose a simple yet effective method for scaling variability in 3D pedestrian assets for synt"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our findings demonstrate that controlled synthetic diversification improves robustness in 2D detection while revealing the sensitivity of 3D perception models to geometric domain gaps.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That automatically mapping StyleGAN2-generated textures onto 3D meshes produces appearance variations that are realistic enough to improve robustness without introducing mapping artifacts or distribution shifts that undermine the detection gains.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Generative texture synthesis from StyleGAN2 diversifies 3D pedestrian assets from a single base model, improving robustness in 2D object detection while exposing 3D perception models' sensitivity to geometric domain gaps.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Synthesizing diverse facial textures on one 3D pedestrian base asset improves 2D detection robustness but exposes geometric sensitivities in 3D point-cloud models.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"59b54de15a3e58d0264a3c14d1f379f1421bc06de610017688dc3987ea6a9d45"},"source":{"id":"2605.13755","kind":"arxiv","version":1},"verdict":{"id":"7d359241-dd10-4671-99b1-83bd11162beb","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:37:36.458101Z","strongest_claim":"Our findings demonstrate that controlled synthetic diversification improves robustness in 2D detection while revealing the sensitivity of 3D perception models to geometric domain gaps.","one_line_summary":"Generative texture synthesis from StyleGAN2 diversifies 3D pedestrian assets from a single base model, improving robustness in 2D object detection while exposing 3D perception models' sensitivity to geometric domain gaps.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That automatically mapping StyleGAN2-generated textures onto 3D meshes produces appearance variations that are realistic enough to improve robustness without introducing mapping artifacts or distribution shifts that undermine the detection gains.","pith_extraction_headline":"Synthesizing diverse facial textures on one 3D pedestrian base asset improves 2D detection robustness but exposes geometric sensitivities in 3D point-cloud models."},"references":{"count":26,"sample":[{"doi":"","year":2024,"title":"Boosting few-shot detection with large language models and layout-to-image synthesis","work_id":"9a31abb6-b769-4afa-9f76-d5de5b4e15ff","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Ffhq-uv: Normalized facial uv-texture dataset for 3d face reconstruction, 2023","work_id":"3ecbd10b-5b67-451f-8ac5-8e2de420be4b","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Vir- tual kitti 2, 2020","work_id":"552b577f-3178-4a91-9394-ca9226c6815c","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Data augmentation for object detec- tion via controllable diffusion models","work_id":"a9648b9d-a7e9-4da6-9687-f6617ee43b92","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"Virtual worlds as proxy for multi-object tracking anal- ysis, 2016","work_id":"328f0cc1-6e52-47f6-bdf5-c43a0cffbe38","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":26,"snapshot_sha256":"b37c88ca1ad47d5ec2cd340e36134ef6ec5f6b23415fe701b1a804930baa9ba4","internal_anchors":2},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}