Results

Quantitative evaluation of our motion-aware video editing pipeline comparing SAM2 and YOLO-based mask propagation strategies.

Mask Quality Metrics

IoU vs Frame Index

Dice Coefficient vs Frame Index

SSIM vs Frame Index

Mask Area Comparison

Video Gallery

Object Replacement: Cat to Drone

Replacing a walking cat with a drone using the full pipeline.

Final Result

End-to-end pipeline output with motion-aware warping and compositing.

Bear Clip Baseline

Baseline clip used for segmentation and inpainting evaluation.

Humans Video Preview

Non-rigid motion test case with articulated human subjects.

Key Findings

0.91
Mean IoU

SAM2 vs YOLO mask overlap across 60 frames

0.95
Mean Dice

High mask agreement between methods

0.98
Mean SSIM

Near-perfect structural similarity in masked regions

~15K
Avg Mask Area

SAM2 produces slightly tighter masks than YOLO

Object Insertion Parameters

Per-Frame Insertion Parameters (ninjas.csv)