● Runs the real numpy library in your browser

Dynamic Object Removal Playground

Powered by Pyodide: the exact dynamic_object_removal.py from this repo runs client-side. No install, no GPU, no upload — your point cloud never leaves your browser.

Booting Python runtime…

Box mode: crop points inside detected 3D boxes (needs boxes). Drop your own scan below.

Range margin 0.50 m See-through votes 2 Surface revert max 2 H resolution 1.0° V resolution 1.0°

Voxel size 0.30 m Min hits 3

Drop your own LiDAR scan
.pcd · .bin (KITTI) · .xyz · .csv · .npy
optional: drop a boxes .json too

Box margin 0.05 m

Total points

—

Removed

—

Removed %

—

Compute time

—

kept (static) removed (dynamic)

show removed points show detection boxes auto-rotate

Point size 1.0

Box removal needs detection boxes (the sample ships with one). Drop a {"objects":[{"center":[x,y,z],"size":[l,w,h],"yaw":r}]} JSON next to your scan to remove your own. Feather/AV2 input needs pyarrow and only works in the CLI.

Sample scenes: Argoverse 2 (64-beam, 95k pts) and nuScenes mini (32-beam, coarser range image). Shareable URLs preserve mode, preset, and algorithm knobs.

← back to all demos · source on GitHub

drag-orbit · scroll-zoom

Link copied