update

Dockerfile

@@ -12,26 +12,5 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-ARG FROM_IMAGE_NAME=nvcr.io/nvidia/pytorch:20.06-py3
-FROM ${FROM_IMAGE_NAME}
+FROM=image.sourcefind.cn:5000/dcu/admin/base/custom:mlperf-resnet50-mpirun-latest
 
-# Install dependencies for system configuration logger
-RUN apt-get update \
- && apt-get install -y --no-install-recommends \
-        infiniband-diags \
-        pciutils \
- && rm -rf /var/lib/apt/lists/*
-
-# Install Python dependencies
-WORKDIR /workspace/object_detection
-
-COPY requirements.txt .
-RUN pip install --no-cache-dir https://github.com/mlperf/logging/archive/9ea0afa.zip \
- && pip install --no-cache-dir -r requirements.txt
-
-# Copy detectron code and build
-COPY . .
-RUN pip install -e .
-RUN /opt/conda/bin/conda install -y numpy==1.17.4
-ENV OMP_NUM_THREADS=1
-ENV OPENCV_FOR_THREADS_NUM=1

README.md

@@ -29,6 +29,8 @@ MaskRCNN通过使用RoI Align算法和Mask Head网络来实现像素级别的物
 
 ## 环境配置
 
+**Docker (方法一)**
+
 提供[光源](https://www.sourcefind.cn/#/service-details)拉取的训练的docker镜像：
 
     docker pull image.sourcefind.cn:5000/dcu/admin/base/custom:mlperf-maskrcnn-latest
@@ -37,6 +39,14 @@ MaskRCNN通过使用RoI Align算法和Mask Head网络来实现像素级别的物
     # <Container Path>容器映射路径
     docker run -it --name mlperf_maskrcnn --shm-size=32G  --device=/dev/kfd --device=/dev/dri/ --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --ulimit memlock=-1:-1 --ipc=host --network host --group-add video -v <Host Path>:<Container Path> <Image ID> /bin/bash
 
+**Dockerfile (方法二)**
+
+    docker build --no-cache -t mlperf_maskrcnn:latest
+    docker run -it --name mlperf_maskrcnn --shm-size=32G  --device=/dev/kfd --device=/dev/dri/ --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --ulimit memlock=-1:-1 --ipc=host --network host --group-add video -v <Host Path>:<Container Path> <Image ID> /bin/bash
+    # <Image ID>用上面拉取docker镜像的ID替换
+    # <Host Path>主机端路径
+    # <Container Path>容器映射路径
+
 镜像版本依赖：
 
 * DTK驱动：dtk22.10
@@ -74,6 +84,25 @@ COCO数据集的广泛使用使得其成为计算机视觉领域中的一个标
     
     #To speed up loading of coco annotations during training, the annotations can be pickled since #unpickling is faster than loading a json.
         python3 pickle_coco_annotations.py --root <path/to/detectron2/dataset/dir> --ann_file <path/to/coco/annotation/file> --pickle_output_file <path/to/pickled/output/file>
+数据集的目录结构如下：
+
+```
+├── images 
+│   ├── train2017
+│   ├── val2017
+│   ├── test2017
+├── labels
+│   ├── train2017
+│   ├── val2017
+├── annotations
+│   ├── instances_val2017.json
+├── LICENSE
+├── README.txt 
+├── test-dev2017.txt
+├── train2017.txt
+├── val2017.txt
+```
+
 ## 训练
 
 ### 单机8卡
@@ -110,7 +139,7 @@ bash sbatch.sh
 ## 源码仓库及问题反馈
 
 * https://developer.hpccube.com/codes/modelzoo/mlperf_maskrcnn_pytorch
-## 参考
+## 参考资料
 * https://mlcommons.org/en/
 * https://github.com/mlcommons
 * https://github.com/mlcommons/training_results_v0.7/tree/master/NVIDIA/benchmarks/maskrcnn/implementations/pytorch

model.properties

@@ -5,7 +5,7 @@ modelName=mlperf_maskrcnn_pytorch
 # 模型描述
 modelDescription=MaskRCNN是一种基于Faster R-CNN的物体检测和实例分割模型
 # 应用场景
-appScenario=训练,CV,零售,广媒,医疗,交通
+appScenario=训练,图像分类,零售,广媒,医疗,交通
 # 框架类型
 frameType=PyTorch