Andreas Geiger Courses

Constraining 3D Fields for Reconstruction and View Synthesis

Regularizing neural radiance fields with geometric cues and tensorial representations enables sparse-view 3D reconstruction and novel view synthesis.

• YouTube
• 27 minutes
• Self-Paced
• Free Video

Learning Robust Policies for Self-Driving

Explores learned end-to-end driving policies using camera–LiDAR fusion, object-level planning, and adversarial scenarios for robust autonomous driving.

• YouTube
• 30 minutes
• Self-Paced
• Free Video

Driving with Attention

Explores attention-based fusion of camera and LiDAR representations, using global scene context and bird’s-eye-view semantics to improve end-to-end autonomous driving.

• YouTube
• 33 minutes
• Self-Paced
• Free Video

Towards Animatable Human Avatars

Learn how neural implicit shapes and meta-learning enable animatable clothed human avatars from deformed observations and few depth images.

• YouTube
• 28 minutes
• Self-Paced
• Free Video

Segmenting and Tracking Every Pixel

Explore video panoptic segmentation, a dense computer-vision task that assigns semantic classes and track identities to every pixel across long real-world video sequences.

• YouTube
• 34 minutes
• Self-Paced
• Free Video

Generative Neural Scene Representations for 3D-Aware Image Synthesis

A research talk on neural radiance field models that synthesize controllable 3D-aware images from unstructured, unposed image collections.

• YouTube
• 26 minutes
• Self-Paced
• Free Video

KITTI-360 - A Novel Dataset and Benchmarks for Urban Scene Understanding in 2D and 3D

A presentation of KITTI-360, a multimodal urban driving dataset with 2D and 3D annotations and benchmarks for scene understanding, novel view synthesis, and semantic SLAM.

• YouTube
• 23 minutes
• Self-Paced
• Free Video

Implicit Neural Scene Representations

Explore how implicit neural representations encode scenes as neural networks and enable neural rendering and 3D reconstruction from sparse visual observations.

• YouTube
• 57 minutes
• Self-Paced
• Free Video

Neural Radiance Fields for View Synthesis

Explores Neural Radiance Fields for synthesizing photorealistic novel views from sparse images using continuous neural scene representations and volumetric rendering.

• YouTube
• 49 minutes
• Self-Paced
• Free Video

Learning 3D Reconstruction in Function Space - Long Version

Explores neural implicit representations that reconstruct 3D geometry, appearance, and motion from images in continuous function space.

• YouTube
• 43 minutes
• Self-Paced
• Free Video

Estimating Human Motion: Past, Present, and Future

A history of human motion estimation, from early computer-vision models to 3D body reconstruction and human-object interaction.

• YouTube
• 1 hour 31 minutes
• Self-Paced
• Free Video