Skip to content

Latest commit

 

History

History
116 lines (92 loc) · 4.52 KB

README.md

File metadata and controls

116 lines (92 loc) · 4.52 KB
Raw

Mask R-CNN Inference

Description

This document has instructions for running Mask R-CNN inference.

Datasets

Download the 2017 COCO dataset using the download_dataset.sh script. Export the DATASET_DIR environment variable to specify the directory where the dataset will be downloaded. This environment variable will be used again when running quickstart scripts.

cd <path to your clone of the model zoo>/quickstart/object_detection/pytorch/maskrcnn/inference/cpu
export DATASET_DIR=<directory where the dataset will be saved>
bash download_dataset.sh

Quick Start Scripts

DataType mode Throughput Latency Accuracy
FP32 imperative bash inference_throughput.sh fp32 imperative bash inference_realtime.sh fp32 imperative bash accuracy.sh fp32 imperative
BF16 imperative bash inference_throughput.sh bf16 imperative bash inference_realtime.sh bf16 imperative bash accuracy.sh bf16 imperative
BF32 imperative bash inference_throughput.sh bf32 imperative bash inference_realtime.sh bf32 imperative bash accuracy.sh bf32 imperative
FP32 jit bash inference_throughput.sh fp32 jit bash inference_realtime.sh fp32 jit bash accuracy.sh fp32 jit
BF16 jit bash inference_throughput.sh bf16 jit bash inference_realtime.sh bf16 jit bash accuracy.sh bf16 jit
BF32 jit bash inference_throughput.sh bf32 jit bash inference_realtime.sh bf32 jit bash accuracy.sh bf32 jit

Follow the instructions to setup your bare metal environment on either Linux or Windows systems. Once all the setup is done, the Model Zoo can be used to run a quickstart script. Ensure that you have a clone of the Model Zoo Github repository.

git clone https://github.com/IntelAI/models.git

Run on Linux

Follow link to install Miniconda and build Pytorch, IPEX, TorchVison and Jemalloc.

  • Install dependencies

    pip install yacs opencv-python pycocotools defusedxml cityscapesscripts
conda install intel-openmp

  • Install model

    cd models/object_detection/pytorch/maskrcnn/maskrcnn-benchmark/
python setup.py develop

  • Download pretrained model

    cd <path to your clone of the model zoo>/quickstart/object_detection/pytorch/maskrcnn/inference/cpu
export CHECKPOINT_DIR=<directory where the pretrained model will be saved>
bash download_model.sh

  • Set Jemalloc Preload for better performance

    After Jemalloc setup, set the following environment variables.

    export LD_PRELOAD="<path to the jemalloc directory>/lib/libjemalloc.so":$LD_PRELOAD
export MALLOC_CONF="oversize_threshold:1,background_thread:true,metadata_thp:auto,dirty_decay_ms:9000000000,muzzy_decay_ms:9000000000"

  • Set IOMP preload for better performance

    IOMP should be installed in your conda env. Set the following environment variables.

    export LD_PRELOAD=<path to the intel-openmp directory>/lib/libiomp5.so:$LD_PRELOAD

  • Set ENV to use AMX if you are using SPR

    export DNNL_MAX_CPU_ISA=AVX512_CORE_AMX

  • Run the model:

    cd models

# Set environment variables
export DATASET_DIR=<path to the COCO dataset>
export CHECKPOINT_DIR=<path to the downloaded pretrained model>
export OUTPUT_DIR=<path to an output directory>
export MODE=<set to 'jit' or 'imperative'>

# Run a quickstart script (for example, FP32 batch inference jit)
bash quickstart/object_detection/pytorch/maskrcnn/inference/cpu/inference_throughput.sh fp32 jit

Run on Windows

If not already setup, please follow instructions for environment setup on Windows.

  • Install dependencies

    pip install yacs opencv-python pycocotools defusedxml cityscapesscripts
conda install intel-openmp

  • Using Windows CMD.exe, run:

    cd models

# Env vars
set DATASET_DIR=<path to the COCO dataset>
set CHECKPOINT_DIR=<path to the downloaded pretrained model>
set OUTPUT_DIR=<path to the directory where log files will be written>
set MODE=<set to 'jit' or 'imperative'>

#Run a quickstart script for fp32 precision(FP32 inference_realtime or inference_throughput or accuracy)
bash quickstart\object_detection\pytorch\maskrcnn\inference\cpu\batch_inference_baremetal.sh fp32 jit

License

LICENSE