|
| 1 | + |
| 2 | +# 从训练到推理部署工具链测试方法介绍 |
| 3 | + |
| 4 | +test_model.sh和params.txt文件配合使用,完成分割模型从训练到预测的流程测试。 |
| 5 | + |
| 6 | +# 安装依赖 |
| 7 | +- 安装PaddlePaddle >= 2.1.2 |
| 8 | +- 安装PaddleSeg依赖 |
| 9 | + ``` |
| 10 | + pip3 install -r ../requirements.txt |
| 11 | + ``` |
| 12 | +- 安装autolog |
| 13 | + ``` |
| 14 | + git clone https://github.com/LDOUBLEV/AutoLog |
| 15 | + cd AutoLog |
| 16 | + pip3 install -r requirements.txt |
| 17 | + python3 setup.py bdist_wheel |
| 18 | + pip3 install ./dist/auto_log-1.0.0-py3-none-any.whl |
| 19 | + cd ../ |
| 20 | + ``` |
| 21 | +
|
| 22 | +# 目录介绍 |
| 23 | +
|
| 24 | +```bash |
| 25 | +tests/ |
| 26 | +├── hrnetw18_small_v1_params.txt # 测试分割模型的参数配置文件 |
| 27 | +└── fcn_hrnetw18_small_v1_humanseg_192x192_mini_supervisely.yml # 测试分割模型的config配置文件 |
| 28 | +└── prepare.sh # 完成test_model.sh运行所需要的数据和模型下载 |
| 29 | +└── test_model.sh # 测试主程序 |
| 30 | +``` |
| 31 | + |
| 32 | +# 使用方法 |
| 33 | + |
| 34 | +test_model.sh包含四种运行模式,每种模式的运行数据不同,分别用于测试速度和精度,分别是: |
| 35 | + |
| 36 | +- 模式1:lite_train_infer,使用少量数据训练,用于快速验证训练到预测的走通流程,不验证精度和速度; |
| 37 | +```shell |
| 38 | +bash test/prepare.sh ./tests/hrnetw18_small_v1_params.txt 'lite_train_infer' |
| 39 | +bash tests/test_model.sh ./tests/hrnetw18_small_v1_params.txt 'lite_train_infer' |
| 40 | +``` |
| 41 | + |
| 42 | +- 模式2:whole_infer,使用少量数据训练,一定量数据预测,用于验证训练后的模型执行预测,预测速度是否合理; |
| 43 | +```shell |
| 44 | +bash tests/prepare.sh ./tests/hrnetw18_small_v1_params.txt 'whole_infer' |
| 45 | +bash tests/test_model.sh ./tests/hrnetw18_small_v1_params.txt 'whole_infer' |
| 46 | +``` |
| 47 | + |
| 48 | +- 模式3:infer 不训练,全量数据预测,走通开源模型评估、动转静,检查inference model预测时间和精度; |
| 49 | +```shell |
| 50 | +bash tests/prepare.sh ./tests/hrnetw18_small_v1_params.txt 'infer' |
| 51 | +# 用法1: |
| 52 | +bash tests/test_model.sh ./tests/hrnetw18_small_v1_params.txt 'infer' |
| 53 | +# 用法2: 指定GPU卡预测,第三个传入参数为GPU卡号 |
| 54 | +bash tests/test_model.sh ./tests/hrnetw18_small_v1_params.txt 'infer' '1' |
| 55 | +``` |
| 56 | + |
| 57 | +- 模式4:whole_train_infer , CE: 全量数据训练,全量数据预测,验证模型训练精度,预测精度,预测速度; |
| 58 | +```shell |
| 59 | +bash tests/prepare.sh ./tests/hrnetw18_small_v1_params.txt 'whole_train_infer' |
| 60 | +bash tests/test_model.sh ./tests/hrnetw18_small_v1_params.txt 'whole_train_infer' |
| 61 | +``` |
| 62 | + |
| 63 | + |
| 64 | +# 日志输出 |
| 65 | +最终在```tests/output```目录下生成.log后缀的日志文件 |
0 commit comments