Adrián López Rodríguez
Computer vision engineer · PhD, Imperial College London · Barcelona
I train, evaluate and deploy vision models, especially when the data you have differs from the data you need: unusual scenes, synthetic training data, noisy sensors. My work covers object detection, depth and 3D reconstruction, and 3D humans.
Work
- Face redaction in mirrors and shower glassA fine-tuned detector that found 14 of 14 reflected faces; YuNet found 10. 2026.
- Synthetic-to-real LiDAR depth completionACCV 2020 Best Student Paper, IJCV 2022, US patent.
- OoD-Pose: camera pose regression trained on synthetic views3DV 2022, joint first author. Renders new views where training poses are missing.
- Monocular depth estimation across domainsBMVC 2020 oral, IJCV 2022.
- Object detection that transfers to new image stylesBMVC 2019. Up to +9.6 mAP over the prior state of the art.
- InstantGeoAvatar: 3D avatars from monocular videoCo-author, 2024. Clothed human geometry in under 10 minutes of training.