KO | EN

GitLite

Minimalist Open Source Repository Search Engine

gitlite — search

#dsh #dsh-plugin #developer-tools #ai-agent #claude-code #codex #typescript #ai-agents #deepseek-harness #mcp #cli #rust

MPCF--3d_object_detection

★ 34 Open GitHub ↗

We propose a method MPCF. This work was published in IEEE TCSVT at 2026.07.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

<div align="center">
   
# MPCF: Multi-Phase Consolidated Fusion for Multi-Modal 3D Object Detection with Pseudo Point Cloud
   
[Pan Gao](https://pangao-1.github.io/) <sup>1, </sup>,
[Ping Zhang](https://github.com/ELOESZHANG) <sup>1,✉* </sup>,
[`Paper`](https://ieeexplore.ieee.org/abstract/document/11398352).
<br>
<sup>1</sup> UESTC
<br>
</div>

We present [`MPCF`](https://doi.org/10.1109/TCSVT.2026.3665922) (MPCF: Multi-Phase Consolidated Fusion for Multi-Modal 3D Object Detection with Pseudo Point Cloud) for point cloud and pseduo point cloud. [`Paper`](https://ieeexplore.ieee.org/abstract/document/11398352) in IEEE TCSVT. You can click the PDF button to download the full text.

![](https://raw.githubusercontent.com/ELOESZHANG/MPCF--3d_object_detection/main/MPCF.png)

This is the official implementation of [**MPCF**],[`Paper`](https://ieeexplore.ieee.org/abstract/document/11398352), built on [`SFD`](https://github.com/LittlePey/SFD) and [`OpenPCDet`](https://github.com/open-mmlab/OpenPCDet)

## 🔥Highlights

* **Strong performance**. MPCF achieves **SOTA** performance on kitti test set for single-use data. [`KITTI Benchmark`](https://www.cvlibs.net/datasets/kitti/eval_object.php?obj_benchmark=3d)💪

* **More friendly**. MPCF utilizes less than **7 GB** during training and  **~3 GB** during inference. (i.e., RTX 3090, RTX4090 are enough to train our MPCF). 😀

### Model Zoo
We release the model based on KITTI Dataset.

* All models are trained with 1 RTX-3090 or 4 RTX-4090 GPUs and are available for download.

* As for KITTI validation set, the models are trained with train split (3712 samples).

* As for KITTI test set, please use slightly score (~0.5) threshold and train the models on all training data to achieve a desirable performance.

|                                             |Modality|GPU memory| Easy | Mod. | Hard  | download | 
|---------------------------------------------|----------:|----------:|:-------:|:-------:|:-------:|:---------:|
| [mpcf-val](tools/cfgs/kitti_models/mpcf.yaml)|LiDAR+RGB|~7GB (train)/~3 GB(val) |95.97 |89.67| 86.89| [google](https://drive.google.com/file/d/1AT2sthr0YhI5ZurtwLghpp60baefC1C_/view?usp=sharing) / [baidu](https://pan.baidu.com/s/1A_A4OY8Kq3xsPLIj9yDNBA?pwd=1234) | 
| [mpcf-test](tools/cfgs/kitti_models/mpcf_can.yaml)|LiDAR+RGB |~7GB (train)/~3 GB(val)| 92.46 |85.50 |80.69 | [`KITTI Benchmark`](https://www.cvlibs.net/datasets/kitti/eval_object.php?obj_benchmark=3d) |

### Installation
1.  Prepare for the running environment.

You can  follow the installation steps in [`OpenPCDet`](https://github.com/open-mmlab/OpenPCDet). We use 1 RTX-3090 or 4 RTX-4090 GPUs to train our MPCF.

2. Prepare for the data.  
    
    The dataset is follow [`SFD`](https://github.com/LittlePey/SFD). Anyway, you should have your dataset as follows:

```
    MPCF
    ├── data
    │   ├── kitti_pseudo
    │   │   │── ImageSets
    │   │   │── training
    │   │   │   ├──calib & velodyne & label_2 & image_2 & (optional: planes) & depth_dense_twise & depth_pseudo_rgbseguv_twise
    │   │   │── testing
    │   │   │   ├──calib & velodyne & image_2 & depth_dense_twise & depth_pseudo_rgbseguv_twise
    │   │   │── gt_database
    │   │   │── gt_database_pseudo_seguv
    │   │   │── kitti_dbinfos_train_custom_seguv.pkl
    │   │   │── kitti_infos_test.pkl
    │   │   │── kitti_infos_train.pkl
    │   │   │── kitti_infos_trainval.pkl
    │   │   │── kitti_infos_val.pkl
    ├── pcdet
    ├── tools
    ```
    .

3. Setup.

```
    conda create -n MPCF_env python=3.8
    conda activate MPCF_env
    
     pip install torch==1.10.1+cu113 torchvision==0.11.2+cu113 torchaudio==0.10.1+cu113 -f https://download.pytorch.org/whl/torch_stable.html
    (or: pip install torch==2.0.0+cu117 torchvision==0.15.1+cu117 torchaudio==2.0.1+cu117 -f https://download.pytorch.org/whl/torch_stable.html )
    pip install -r requirements.txt
    pip install spconv-cu113 (or spconv-cu116)

cd MPCF
    python setup.py develop
    
    cd pcdet/ops/iou3d/cuda_op
    python setup.py develop

```

### Getting Started

You can find the training and testing commands in tools/GP_run.sh

0. Creat kitti_pkl and GT

```
    python -m pcdet.datasets.kitti.kitti_dataset_custom create_kitti_infos ../tools/cfgs/dataset_configs/kitti_dataset_custom.yaml
    ```
    
1. Training. (We recommend running on single GPU, and our optimal model was trained using just 1 GPU.)

For single GPU 
    ```
    cd tools
    python train.py --gpu_id 0 --workers 0 --cfg_file cfgs/kitti_models/mpcf.yaml \
     --batch_size 1 --epochs 60 --max_ckpt_save_num 25 --fix_random_seed
    ```
    
    For 4 GPUs
    ```
    cd tools
    python -m torch.distributed.launch --nnodes 1 --nproc_per_node=4 --master_port 25511 train.py \
     --gpu_id 0,1,2,3 --launch 'pytorch' --workers 4 \
     --batch_size 4 --cfg_file cfgs/kitti_models/mpcf.yaml  --tcp_port 61000 \
     --epochs 60 --max_ckpt_save_num 30 --fix_random_seed
    ```

2. Evaluation.

```
    cd tools
    python test.py --gpu_id 1 --workers 4 --cfg_file cfgs/kitti_models/mpcf_test.yaml --batch_size 1 \
     --ckpt ../output/kitti_models/mpcf/default/ckpt/checkpoint_epoch_57.pth #--save_to_file 
    ```
    
## License

This code is released under the [Apache 2.0 license](LICENSE).
    
## Acknowledgements
We thank these great works and open-source repositories:
[OpenPCDet](https://github.com/open-mmlab/OpenPCDet), [SFD](https://github.com/LittlePey/SFD), and [Voxel-RCNN](https://github.com/djiajunustc/Voxel-R-CNN).

## Citation

MPCF--3d_object_detection

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README