ContactPose: 2.9 Million RGB-D Images of Hand-Object Interaction

ContactPose contains 2.9 million RGB-D grasp images paired with hand and object pose data, developed by Facebook Research. The collection includes synchronized contact maps and MANO hand model parameters for various hand-object interactions. It serves as a benchmark for understanding how humans manipulate objects using visual and tactile data.

Use Cases

• Predicting contact maps from RGB images to identify tactile interaction points
• Estimating MANO hand model parameters during complex object manipulation
• Training robotic grasping algorithms using the 2.9 million RGB-D image frames

Strengths

• 2.9 million RGB-D images
• Includes MANO hand model parameters
• Synchronized hand and object pose labels

Limitations

• Limited to the specific set of objects used during the capture session
• Potential sensor noise inherent in RGB-D depth maps

Provenance

Source

Facebook Research

Collection Method

Sensor-based capture using RGB-D cameras and pose tracking systems

Freshness

Last updated May 2025.