Browser-based 3D LiDAR visualization — load official dataset files from disk, view colored point clouds and 3D bounding boxes with zero install. Works offline after the first load; supports immersive VR on Meta Quest and other WebXR browsers.

Live app: https://ankk98.github.io/parmanu-viz/

1. Open the app:

   • Online: https://ankk98.github.io/parmanu-viz/
   • Local: Just open index.html

2. Choose a Dataset.

3. Pick the frame files (see Datasets).

4. Click Visualize.

Use mouse orbit / zoom / pan, or the panel buttons and keyboard shortcuts (Normal (desktop) controls).

VR needs HTTPS (GitHub Pages or localhost). file:// will not enable WebXR.

1. On a PC, copy one frame’s files into a folder (e.g. parmanu/):
   • KITTI: 000001.bin, 000001.txt (label_2), 000001.txt (calib)
   • SiT: {frame}.pcd and, if using labels, matching label_3d + ego_trajectory .txt files
2. Transfer the folder to the headset (USB → Internal storage/Download/parmanu/, or adb push below).
3. On the headset, open Quest Browser and go to https://ankk98.github.io/parmanu-viz/.
4. Select the dataset, pick files from Download/parmanu/, tap Visualize, then Enter VR.
5. Move with controllers (Immersive VR controls).

Same Wi‑Fi (developer): serve from your PC and open http://<your-pc-ip>:8000/ on the Quest:

python -m http.server 8000 --bind 0.0.0.0

ADB (optional):

adb push 000001.bin /sdcard/Download/parmanu/
adb push 000001.txt /sdcard/Download/parmanu/   # repeat for label and calib

If “Enter VR” does not appear: use Quest Browser (not a non-XR browser), ensure the URL is HTTPS, enable WebXR in chrome://flags, reload, and visualize a frame first (the button stays disabled until a scene is loaded).

Standard z-up orbit camera around the LiDAR frame. Intensity-colored points (when the dataset provides a fourth channel) and wireframe 3D boxes.

WebXR immersive-vr in Quest Browser and other XR-capable browsers. Same point cloud and boxes as desktop, rendered in an xr-standard rig (y-up in-headset; content rotated for LiDAR alignment). Large clouds are decimated to 80k points in VR to keep framerate stable.

Comfort tips: prefer teleport over stick-only movement; allow arm room; stand or sit as you prefer.

VR screenshot: In-headset captures are welcome — add images under screenshots/ (e.g. vr-kitti.png) via a pull request.

parmanu-viz reads official raw files via the browser file picker (FileReader). No Python preprocess step, no conversion to .bin for SiT.

Per frame (same 6-digit id, e.g. 000001):

Notes:

• Parses Velodyne .bin as N×4 float32 (x, y, z, reflectance).
• Calibration must include P2, R0_rect, Tr_velo_to_cam.
• 3D boxes are projected to Velodyne using logic ported from the KITTI object devkit / kitti_object_vis.
• Only objects that pass the loader’s 3D validity filter are drawn.

Per frame under {scene}/{sequence}/:

Notes:

• Loads binary_compressed PCD via bundled THREE.PCDLoader (r134). PCD intensity is ignored in v1 (fourth channel set to 0).
• label_3d lines are world-frame; with labels, pick the matching ego trajectory so boxes align with the cloud (same transform as SiT-Dataset visualize_3dbox_on_image.py).
• Skip ego transform: checkbox applies raw xyz + yaw flip only (debug / comparison); default uses official viz math.
• Sequences without label_3d (e.g. Lobby_1, Outdoor_Alley_1, Subway_Entrance_1) work in points-only mode — select .pcd only.
• No KITTI-style rectified-camera pipeline; boxes are LiDAR [x,y,z,l,w,h,yaw] after ego transform.

Static site: plain HTML + classic <script> tags. No npm, no bundler, no ES modules.

Scene graph (WebXR-ready): viewer.js builds _xrRig → camera, and _contentGroup → point clouds (desktop + VR-decimated), box line segments, grid/origin helpers in VR. Desktop uses OrbitControls; VR disables orbit and drives _xrRig from vr.js.

Data flow: file pickers → Loader.loadFrame() → { points: Float32Array, boxes: [{ type, corners }] } → viewer.loadScene().

Deployment: GitHub Actions uploads the repo root to GitHub Pages (.nojekyll at root). See .github/workflows/deploy-pages.yml.

1. Add js/datasets/<name>.js with a Loader implementing loadFrame() → { points, boxes }, plus createExplorer for file pickers.
2. Register: DatasetRegistry.register('<id>', { name, fileHint, Loader, createExplorer }).
3. Add <script src="js/datasets/<name>.js"></script> in index.html after registry.js.
4. Add a file panel block in index.html and hook visibility in app.js if the UI differs from existing datasets.

The Dataset dropdown updates from the registry automatically.

parmanu (Sanskrit: atom / particle) — visualization should stay out of the way:

• Zero install — open a browser, pick files, visualize.
• Offline-first — works without a backend after the first load.
• No mandatory conversion — use official dataset files; ego/calib handled in loaders.
• Extensible — new datasets are one JS file + registry entry.

Issues and pull requests are welcome. No build step: test by serving the repo and loading sample frames.

• New datasets — loader under js/datasets/ (see Add another dataset).
• Bugs / UX — viewer, pickers, calibration or ego edge cases.
• Docs — dataset notes, screenshots (including VR captures).

For larger changes, open an issue first to align on approach.

MIT — see LICENSE.