150 lines
3.2 KiB
Markdown
150 lines
3.2 KiB
Markdown
|
# 快速开始：双ROI融合模型测试
|
|||
|
|
|
|||
|
|
## 5分钟快速上手
|
|||
|
|
|
|||
|
|
### 1. 检查依赖
|
|||
|
|
|
|||
|
|
```bash
|
|||
|
|
# 确保在项目根目录
|
|||
|
|
cd /path/to/yolov5-3d
|
|||
|
|
|
|||
|
|
# 检查Python包
|
|||
|
|
python -c "import numpy, cv2, torch; print('基础依赖OK')"
|
|||
|
|
|
|||
|
|
# 如果测试ONNX模型，检查onnxruntime
|
|||
|
|
python -c "import onnxruntime; print('ONNX Runtime已安装')"
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
### 2. 准备模型和数据
|
|||
|
|
|
|||
|
|
确保以下文件存在：
|
|||
|
|
- 模型文件：`release/yolov5s-30w/merged_model.onnx` 或 `.torchscript`
|
|||
|
|
- 测试数据：视频bin文件或图像目录
|
|||
|
|
|
|||
|
|
### 3. 运行测试
|
|||
|
|
|
|||
|
|
#### 方式一：使用交互式脚本（推荐）
|
|||
|
|
|
|||
|
|
```bash
|
|||
|
|
./eval_tools/test_merged_model.sh
|
|||
|
|
# 然后按提示选择测试模式（1-7）
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
#### 方式二：直接命令
|
|||
|
|
|
|||
|
|
```bash
|
|||
|
|
# ONNX模型
|
|||
|
|
python eval_tools/test_val_merged_model.py \
|
|||
|
|
--source /data1/dongying/Mono3d/G1M3/eval_dataset/20251210222153/sigmastar.1/camera4.bin \
|
|||
|
|
--weights release/yolov5s-30w/merged_model.onnx \
|
|||
|
|
--model-type onnx
|
|||
|
|
|
|||
|
|
# TorchScript模型
|
|||
|
|
python eval_tools/test_val_merged_model.py \
|
|||
|
|
--source /data1/dongying/Mono3d/G1M3/eval_dataset/20251210222153/sigmastar.1/camera4.bin \
|
|||
|
|
--weights release/yolov5s-30w/merged_model.torchscript \
|
|||
|
|
--model-type torchscript \
|
|||
|
|
--device cuda
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
### 4. 查看结果
|
|||
|
|
|
|||
|
|
```bash
|
|||
|
|
# 结果保存在
|
|||
|
|
ls runs/val_viz_merged/exp/
|
|||
|
|
|
|||
|
|
# 用图像查看器打开
|
|||
|
|
eog runs/val_viz_merged/exp/000000_*.jpg
|
|||
|
|
# 或
|
|||
|
|
display runs/val_viz_merged/exp/000000_*.jpg
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
## 常用配置
|
|||
|
|
|
|||
|
|
### 提高检测精度（减少误检）
|
|||
|
|
|
|||
|
|
```bash
|
|||
|
|
python eval_tools/test_val_merged_model.py \
|
|||
|
|
--source your/data/path \
|
|||
|
|
--weights your/model.onnx \
|
|||
|
|
--model-type onnx \
|
|||
|
|
--conf-thres 0.4 \
|
|||
|
|
--iou-thres 0.5
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
### 获取更多检测（增加召回）
|
|||
|
|
|
|||
|
|
```bash
|
|||
|
|
python eval_tools/test_val_merged_model.py \
|
|||
|
|
--source your/data/path \
|
|||
|
|
--weights your/model.onnx \
|
|||
|
|
--model-type onnx \
|
|||
|
|
--conf-thres 0.15 \
|
|||
|
|
--iou-thres 0.6
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
## 文件说明
|
|||
|
|
|
|||
|
|
| 文件 | 说明 |
|
|||
|
|
|------|------|
|
|||
|
|
| `test_val_merged_model.py` | 主测试脚本 |
|
|||
|
|
| `test_merged_model.sh` | 交互式测试脚本（推荐） |
|
|||
|
|
| `README_MERGED_MODEL.md` | 完整文档（详细说明） |
|
|||
|
|
| `QUICKSTART_MERGED.md` | 本文件（快速入门） |
|
|||
|
|
|
|||
|
|
## 故障排查
|
|||
|
|
|
|||
|
|
### 问题1：找不到模块
|
|||
|
|
|
|||
|
|
```
|
|||
|
|
ModuleNotFoundError: No module named 'utils'
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
**解决**：确保在项目根目录运行
|
|||
|
|
```bash
|
|||
|
|
cd /deeplearning_team/ydong/dongying/projects/yolov5-3d
|
|||
|
|
python eval_tools/test_val_merged_model.py ...
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
### 问题2：ONNX Runtime错误
|
|||
|
|
|
|||
|
|
```
|
|||
|
|
ImportError: onnxruntime not installed
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
**解决**：
|
|||
|
|
```bash
|
|||
|
|
pip install onnxruntime-gpu # GPU版本
|
|||
|
|
# 或
|
|||
|
|
pip install onnxruntime # CPU版本
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
### 问题3：CUDA内存不足
|
|||
|
|
|
|||
|
|
```
|
|||
|
|
RuntimeError: CUDA out of memory
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
**解决**：使用CPU推理
|
|||
|
|
```bash
|
|||
|
|
# TorchScript模型
|
|||
|
|
python eval_tools/test_val_merged_model.py \
|
|||
|
|
--model-type torchscript \
|
|||
|
|
--device cpu \
|
|||
|
|
...
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
## 下一步
|
|||
|
|
|
|||
|
|
- 阅读完整文档：`README_MERGED_MODEL.md`
|
|||
|
|
- 调整检测参数优化效果
|
|||
|
|
- 修改代码处理更多帧（默认10帧）
|
|||
|
|
- 集成到自己的推理流程
|
|||
|
|
|
|||
|
|
## 技术支持
|
|||
|
|
|
|||
|
|
遇到问题？请检查：
|
|||
|
|
1. 完整文档 `README_MERGED_MODEL.md`
|
|||
|
|
2. 脚本帮助 `python eval_tools/test_val_merged_model.py --help`
|
|||
|
|
3. 交互式脚本选项7查看详细帮助
|