Research

Compressed Map Priors for 3D Perception

Brady Zhou, Philipp Krähenbühl

arXiv 2025

Learnable priors from past traversals.

paper

Cross-view Transformers for Real-Time Map-view Semantic Segmentation

Brady Zhou, Philipp Krähenbühl

CVPR 2022, Oral

Learn map-view embeddings via cross-attention over image features, camera pose, and intrinsics.

paper code

Domain Adaptation through Task Distillation

Brady Zhou, Nimit Kalra, Philipp Krähenbühl

ECCV 2020

Sim2Sim driving by distilling a teacher that uses depth/segmentation as input.

paper code

Learning by Cheating

Dian Chen, Brady Zhou, Vladlen Koltun, Philipp Krähenbühl

CoRL 2019, Spotlight

paper code

End-to-end driving by distilling a privileged map-based policy into a vision-only one.

Does computer vision matter for action?

Brady Zhou, Philipp Krähenbühl, Vladlen Koltun

Science Robotics 2019

Predicting depth and segmentation as intermediate steps improves end-to-end agent performance.

paper code

Don't let your Discriminator be fooled

Brady Zhou, Philipp Krähenbühl

ICLR 2019

Training GANs with adversarial examples smooths the loss landscape and improves generation.

paper code poster

GPU accelerated k-nearest neighbor kernel for sparse feature datasets

Brady Zhou, George Biros

UT MS Research

Accelerated KNN for sparse features by writing some fancy cuda kernels (bitonic sort).

link code

Teaching

• Deep Learning: Online AI MS 2024 (1000+ students), Course Staff; grading infra.
• Deep Learning: Online AI MS 2025 (1000+ students), Course Staff.
• Deep Learning: Online AI MS 2023 (700 students), Course Staff; modernize course + materials.
• Deep Learning: Online AI MS 2022 (400 students), Course Staff.
• Neural Networks, Fall 2020 (200 students), Instructor; lectures.
• Neural Networks, Fall 2019 (100 students), Teaching Assistant.
• Introduction to Programming, Fall 2015 (100 students), Teaching Assistant.

Fun