# pointcloud_segmentation_codebase

**Repository Path**: suyunzzz/pointcloud_segmentation_codebase

## Basic Information

- **Project Name**: pointcloud_segmentation_codebase
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2021-11-19
- **Last Updated**: 2021-12-07

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

<!--
 * @Description: 
 * @Author: suyunzheng
 * @Date: 2021-11-22 22:38:47
 * @LastEditTime: 2021-12-07 22:14:24
 * @LastEditors: maple
-->

# Usage


## train
遵循以下格式：
```
python train_s3dis.py --voxel_size <voxel_size> --gpu_num <1 or 2> --model <model> --log_dir <model-dataset>
```
例如：

```bash
# s3dis
python train_s3dis.py --voxel_size 0.05 --gpu_num 1 --model spvcnn --epoch 600 --dataset s3dis --lr 0.001
```

```bash
# kitti
 python train_kitti.py --voxel_size 0.05 --gpu_num 1 --model spvcnn --dataset kitti --epoch 15 --batch_size 2
```


```
# drinet-s3dis
python train_s3dis_drinet.py --voxel_size 0.2 --gpu_num 1 --model drinet --dataset s3dis --lr 0.001 --batch_size 1 --epoch 600
```


```
# drinet-kitti
python train_kitti_drinet.py --voxel_size 0.2 --gpu_num 1 --model drinet --dataset kitti --epoch 15 --batch_size 2
```
---

# dependence

1. torch-sparse
2. open3d
3. torch1.7
4. ...

# code-base

## step1. process Stanford dataset

>使用voxel=0.1进行下采样

```python
python -m lib.precess_s3dis
```

>参考的是[Minkowski的预处理部分](https://github.com/chrischoy/SpatioTemporalSegmentation/blob/master/lib/datasets/preprocessing/stanford.py)

## setp2. Dataset类

### 1、Stanford
在代码内部会进行`voxel=0.03`的下采样


### 2、SemanticKitti



## step3. model设计


## step4. train


## step5. inference and visualize and dump