---
license: apache-2.0
tags:
  - 3d-object-detection
  - multi-object-tracking
  - gaussian-splatting
  - camera-radar-fusion
  - nuscenes
datasets:
  - nuscenes
---

# GaussianMOT

Joint 3D object detection and tracking on a camera–radar Gaussian BEV.
Code: <https://github.com/KimJunHan/GaussianMOT>

This repository holds the checkpoint reported in the paper (weights only,
end of the 24th training epoch).

| File | Size | SHA-256 |
|---|---|---|
| `gaussianmot_epoch23.ckpt` | 368 MB | `d3e3a5e55265b878420dbbcf1d6868e47350e8c968fb5061e2aad3b88af5dd23` |

## Results (nuScenes val, inference seed 0)

| Detection | | Tracking | |
|---|---|---|---|
| mAP  | 35.0 | AMOTA  | 0.348 |
| NDS  | 42.5 | AMOTP  | 1.31 |
| mATE | 0.539 | Recall | 0.411 |
| mASE | 0.307 | ID-S   | 897 |
| mAOE | 0.681 | | |
| mAVE | 0.511 | | |
| mAAE | 0.465 | | |

6 cameras + 5 radars, 448×800 input, 6.5 FPS on an RTX 3090 (batch size 1).

## Usage

```bash
git clone https://github.com/KimJunHan/GaussianMOT && cd GaussianMOT
hf download kimjunhan/GaussianMOT gaussianmot_epoch23.ckpt --local-dir ckpt
python tools/evaluate.py checkpoint_path=ckpt/gaussianmot_epoch23.ckpt device=cuda
```

The file contains `state_dict` only (no optimizer state); it loads with
`tools/evaluate.py` as is. Densification draws random shifts at inference, so
results depend on `eval_seed` (default 0) in the last digit.