From df2540b9c1f1b69ff2c99ef97a5cf40af6b19334 Mon Sep 17 00:00:00 2001 From: root Date: Tue, 5 Jan 2021 13:56:25 +0800 Subject: [PATCH] maskrcnn and fasterrcnn Dockerfile --- model_zoo/official/cv/faster_rcnn/Dockerfile | 3 +- model_zoo/official/cv/faster_rcnn/README.md | 1 + .../official/cv/faster_rcnn/README_CN.md | 1 + model_zoo/official/cv/maskrcnn/Dockerfile | 3 +- model_zoo/official/cv/maskrcnn/README.md | 3 +- model_zoo/official/cv/maskrcnn/README_CN.md | 70 ++++++++++++++++++- 6 files changed, 75 insertions(+), 6 deletions(-) diff --git a/model_zoo/official/cv/faster_rcnn/Dockerfile b/model_zoo/official/cv/faster_rcnn/Dockerfile index 5ff5b4e8ab..fcb31f207f 100644 --- a/model_zoo/official/cv/faster_rcnn/Dockerfile +++ b/model_zoo/official/cv/faster_rcnn/Dockerfile @@ -1,5 +1,6 @@ -ARG FROM_IMAGE_NAME=ascend-mindspore-arm:20.1.0 +ARG FROM_IMAGE_NAME FROM ${FROM_IMAGE_NAME} +RUN apt install libgl1-mesa-glx -y COPY requirements.txt . RUN pip3.7 install -r requirements.txt diff --git a/model_zoo/official/cv/faster_rcnn/README.md b/model_zoo/official/cv/faster_rcnn/README.md index 0376dece2e..bcb6aaa9b4 100644 --- a/model_zoo/official/cv/faster_rcnn/README.md +++ b/model_zoo/official/cv/faster_rcnn/README.md @@ -5,6 +5,7 @@ - [Dataset](#dataset) - [Environment Requirements](#environment-requirements) - [Quick Start](#quick-start) +- [Run in docker](#Run-in-docker) - [Script Description](#script-description) - [Script and Sample Code](#script-and-sample-code) - [Training Process](#training-process) diff --git a/model_zoo/official/cv/faster_rcnn/README_CN.md b/model_zoo/official/cv/faster_rcnn/README_CN.md index 64d1a15876..a27670b57a 100644 --- a/model_zoo/official/cv/faster_rcnn/README_CN.md +++ b/model_zoo/official/cv/faster_rcnn/README_CN.md @@ -6,6 +6,7 @@ - [数据集](#数据集) - [环境要求](#环境要求) - [快速入门](#快速入门) +- [在docker上运行](#在docker上运行) - [脚本说明](#脚本说明) - [脚本及样例代码](#脚本及样例代码) - [训练过程](#训练过程) diff --git a/model_zoo/official/cv/maskrcnn/Dockerfile b/model_zoo/official/cv/maskrcnn/Dockerfile index 5ff5b4e8ab..fcb31f207f 100644 --- a/model_zoo/official/cv/maskrcnn/Dockerfile +++ b/model_zoo/official/cv/maskrcnn/Dockerfile @@ -1,5 +1,6 @@ -ARG FROM_IMAGE_NAME=ascend-mindspore-arm:20.1.0 +ARG FROM_IMAGE_NAME FROM ${FROM_IMAGE_NAME} +RUN apt install libgl1-mesa-glx -y COPY requirements.txt . RUN pip3.7 install -r requirements.txt diff --git a/model_zoo/official/cv/maskrcnn/README.md b/model_zoo/official/cv/maskrcnn/README.md index 9637b45707..fb0c3465b5 100644 --- a/model_zoo/official/cv/maskrcnn/README.md +++ b/model_zoo/official/cv/maskrcnn/README.md @@ -5,6 +5,7 @@ - [Dataset](#dataset) - [Environment Requirements](#environment-requirements) - [Quick Start](#quick-start) +- [Run in docker](#Run-in-docker) - [Script Description](#script-description) - [Script and Sample Code](#script-and-sample-code) - [Script Parameters](#script-parameters) @@ -397,7 +398,7 @@ bash run_distribute_train.sh [RANK_TABLE_FILE] [PRETRAINED_MODEL] - Notes 1. hccl.json which is specified by RANK_TABLE_FILE is needed when you are running a distribute task. You can generate it by using the [hccl_tools](https://gitee.com/mindspore/mindspore/tree/master/model_zoo/utils/hccl_tools). -2. As for PRETRAINED_MODEL，it should be a trained ResNet50 checkpoint. If not set, the model will be trained from the very beginning. If you need to load Ready-made pretrained FasterRcnn checkpoint, you may make changes to the train.py script as follows. +2. As for PRETRAINED_MODEL，it should be a trained ResNet50 checkpoint. If not set, the model will be trained from the very beginning. If you need to load Ready-made pretrained MaskRcnn checkpoint, you may make changes to the train.py script as follows. ```python # Comment out the following code diff --git a/model_zoo/official/cv/maskrcnn/README_CN.md b/model_zoo/official/cv/maskrcnn/README_CN.md index d4758149e3..391220e18f 100644 --- a/model_zoo/official/cv/maskrcnn/README_CN.md +++ b/model_zoo/official/cv/maskrcnn/README_CN.md @@ -58,6 +58,8 @@ MaskRCNN是一个两级目标检测网络，作为FasterRCNN的扩展模型， - 采用昇腾处理器搭建硬件环境。如需试用昇腾处理器，请发送[申请表](https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/file/other/Ascend%20Model%20Zoo%E4%BD%93%E9%AA%8C%E8%B5%84%E6%BA%90%E7%94%B3%E8%AF%B7%E8%A1%A8.docx)至ascend@huawei.com，审核通过即可获得资源。 - 框架 - [MindSpore](https://gitee.com/mindspore/mindspore) +- 获取基础镜像 + - [Ascend Hub](ascend.huawei.com/ascendhub/#/home) - 如需查看详情，请参见如下资源： - [MindSpore教程](https://www.mindspore.cn/tutorial/training/zh-CN/master/index.html) - [MindSpore Python API](https://www.mindspore.cn/doc/api_python/zh-CN/master/index.html) @@ -134,6 +136,39 @@ pip install mmcv=0.2.14 1. AIR_PATH是在910上使用export脚本导出的模型。 2. ANN_FILE_PATH是推理使用的标注文件。 +# 在docker上运行 + +1. 编译镜像 + +```shell +# 编译镜像 +docker build -t maskrcnn:20.1.0 . --build-arg FROM_IMAGE_NAME=ascend-mindspore-arm:20.1.0 +``` + +2. 启动容器实例 + +```shell +# 启动容器实例 +bash scripts/docker_start.sh maskrcnn:20.1.0 [DATA_DIR] [MODEL_DIR] +``` + +3. 训练 + +```shell +# 单机训练 +bash run_standalone_train.sh [PRETRAINED_CKPT] + +# 分布式训练 +bash run_distribute_train.sh [RANK_TABLE_FILE] [PRETRAINED_CKPT] +``` + +4. 评估 + +```shell +# 评估 +bash run_eval.sh [VALIDATION_JSON_FILE] [CHECKPOINT_PATH] +``` + # 脚本说明 ## 脚本和样例代码 @@ -358,9 +393,38 @@ sh run_standalone_train.sh [PRETRAINED_MODEL] sh run_distribute_train.sh [RANK_TABLE_FILE] [PRETRAINED_MODEL] ``` -> 运行分布式任务时要用到由RANK_TABLE_FILE指定的hccl.json文件。您可使用[hccl_tools](https://gitee.com/mindspore/mindspore/tree/master/model_zoo/utils/hccl_tools)生成该文件。 -> 若不设置PRETRAINED_MODEL，模型将会从头开始训练。暂无预训练模型可用，请持续关注。 -> 本操作涉及处理器内核绑定，需要设置`device_num`及处理器总数。若无需此操作，请删除`scripts/run_distribute_train.sh`中的`taskset`。 +- Notes + +1. 运行分布式任务时要用到由RANK_TABLE_FILE指定的hccl.json文件。您可使用[hccl_tools](https://gitee.com/mindspore/mindspore/tree/master/model_zoo/utils/hccl_tools)生成该文件。 +2. PRETRAINED_MODEL应该是训练好的ResNet50检查点。如果此参数未设置，网络将从头开始训练。如果想要加载训练好的MaskRcnn检查点，需要对train.py作如下修改： + +```python +# Comment out the following code +# load_path = args_opt.pre_trained +# if load_path != "": +# param_dict = load_checkpoint(load_path) +# for item in list(param_dict.keys()): +# if not item.startswith('backbone'): +# param_dict.pop(item) +# load_param_into_net(net, param_dict) + +# Add the following codes after optimizer definition since the FasterRcnn checkpoint includes optimizer parameters： + lr = Tensor(dynamic_lr(config, rank_size=device_num, start_steps=config.pretrain_epoch_size * dataset_size), + mstype.float32) + opt = Momentum(params=net.trainable_params(), learning_rate=lr, momentum=config.momentum, + weight_decay=config.weight_decay, loss_scale=config.loss_scale) + + if load_path != "": + param_dict = load_checkpoint(load_path) + if config.pretrain_epoch_size == 0: + for item in list(param_dict.keys()): + if item in ("global_step", "learning_rate") or "rcnn.cls" in item or "rcnn.mask" in item: + param_dict.pop(item) + load_param_into_net(net, param_dict) + load_param_into_net(opt, param_dict) +``` + +3. 本操作涉及处理器内核绑定，需要设置`device_num`及处理器总数。若无需此操作，请删除`scripts/run_distribute_train.sh`中的`taskset` ### 训练结果