TRI-ML/packnet-sfm

Health v2 · maintenance only

Adoption not part of the score

1274 stars · 246 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Official PyTorch implementation of PackNet and related self-supervised monocular depth estimation methods from Toyota Research Institute's ML team. It provides training and evaluation code for depth and ego-motion estimation from monocular video, including support for non-pinhole cameras via Neural Ray Surfaces.

Use cases

• estimate depth from monocular video
• train a self-supervised depth estimation model
• run monocular depth estimation on KITTI or DDAD
• estimate camera ego-motion from video
• deploy real-time depth estimation with TensorRT
• learn depth and pose on fisheye or catadioptric cameras
• reproduce CVPR 2020 PackNet paper results

When to choose

• you need state-of-the-art self-supervised monocular depth estimation in PyTorch
• you want to reproduce or build on the PackNet or NRS papers
• you work with non-pinhole cameras like fisheye lenses
• you need a research baseline for depth and ego-motion learning

When to avoid

• you need active development or support for new publications - use TRI-ML/vidar instead
• you lack an NVIDIA GPU with at least 6GB of memory
• you need a production-ready depth estimation service rather than research code
• you need stereo or LiDAR-based depth estimation

Facets

library · maturity maintenance

machine-learning deep-learning computer-vision image-processing computer-vision machine-learning autonomous-vehicles deep-learning python monocular-depth-estimation self-supervised-learning pytorch depth-estimation pose-estimation research-code packnet kitti ddad linux docker gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="TRI-ML/packnet-sfm")