Shadow Detection and Texture Segmentation Dataset for Mobile Robots

Four main subfolders contain image sequences for testing shadow detection and texture segmentation algorithms. The dataset includes ground-truthed sequences of textured surfaces with moving shadows, camera travel, a computer-generated 3D scene, and challenging footage from a bipedal robot. Charles C. Newey from the University of Southampton created this dataset for computer vision research.

Use Cases

• Testing shadow detection algorithms based on sequences with an active (moving) camera.
• Evaluating texture segmentation algorithms based on images of various textured surfaces.
• Benchmarking algorithm robustness based on the challenging 'kondo' subfolder containing high noise and poor lighting.
• Training models for shadow removal based on ground-truthed image sequences.

Strengths

• Contains four distinct subfolders ('active', 'artificial', 'kondo', 'static') with different data characteristics.
• Includes ground-truthed image sequences for algorithm evaluation.
• Features a subfolder of challenging real-world data captured from a bipedal robot with complex shadows and noise.

Limitations

• Dataset size, row count, and specific file formats are unknown, limiting suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The 'artificial' subfolder lacks texture in all images, which may limit its utility for texture-specific tasks.

Provenance

Source

Charles C. Newey, University of Southampton

Collection Method

Likely contains captured image sequences and computer-generated scenes, as described.