sarayusapa/3dgs-room-capture

3DGS Room Capture

A trained 3D Gaussian Splatting model of an indoor room capture.

Model Details

• Framework: 3DGS (3D Gaussian Splatting)
• Training Steps: 30,000
• Scene: Indoor room capture with furniture
• Checkpoint: ckpt_30000.pt
• Resolution: Full resolution Gaussian splats

Usage

from threedgrut.render import Renderer

renderer = Renderer.from_checkpoint(
    checkpoint_path="ckpt_30000.pt",
    path="",
    out_dir="./renders"
)

# Render frames
renderer.render_all()

Performance

• Mean PSNR: ~28 dB on validation set
• Inference Speed: ~2.97 ms/frame

Sample Renders

See sample_render.png for a sample rendered frame.

Citation

If you use this model, please cite the original 3DGS paper and the 3DGUT framework.

---

Uploaded with ❤️ via Hugging Face Hub