Tensorflow Object 检测 API 与 GPU 安装

Question

I wanted to try creating an object detection model, so I followed this official tutorial . Everything worked fine. I could use GPU, and no error while I was installing COCO API. As I checked by "Verify the installation" step in the tutorial.

"Verify the installation" tell me to run below code

python -c "import tensorflow as tf;print(tf.reduce_sum(tf.random.normal([1000, 1000])))"

case 1: If I can not use GPU, I would see a print-out similar to the one bellow

2020-06-22 19:20:32.614181: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cudart64_101.dll'; dlerror: cudart64_101.dll not found
2020-06-22 19:20:32.620571: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
2020-06-22 19:20:35.027232: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library nvcuda.dll
2020-06-22 19:20:35.060549: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 0 with properties:
pciBusID: 0000:02:00.0 name: GeForce GTX 1070 Ti computeCapability: 6.1
coreClock: 1.683GHz coreCount: 19 deviceMemorySize: 8.00GiB deviceMemoryBandwidth: 238.66GiB/s
2020-06-22 19:20:35.074967: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cudart64_101.dll'; dlerror: cudart64_101.dll not found
2020-06-22 19:20:35.084458: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cublas64_10.dll'; dlerror: cublas64_10.dll not found
2020-06-22 19:20:35.094112: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cufft64_10.dll'; dlerror: cufft64_10.dll not found
2020-06-22 19:20:35.103571: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'curand64_10.dll'; dlerror: curand64_10.dll not found
2020-06-22 19:20:35.113102: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cusolver64_10.dll'; dlerror: cusolver64_10.dll not found
2020-06-22 19:20:35.123242: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cusparse64_10.dll'; dlerror: cusparse64_10.dll not found
2020-06-22 19:20:35.140987: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2020-06-22 19:20:35.146285: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1598] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...
2020-06-22 19:20:35.162173: I tensorflow/core/platform/cpu_feature_guard.cc:143] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2020-06-22 19:20:35.178588: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x15140db6390 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-06-22 19:20:35.185082: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
2020-06-22 19:20:35.191117: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-06-22 19:20:35.196815: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1108]
tf.Tensor(1620.5817, shape=(), dtype=float32)

case 2: If it can found my GPU, print-out would be like below.

2020-06-22 20:24:31.355541: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2020-06-22 20:24:33.650692: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library nvcuda.dll
2020-06-22 20:24:33.686846: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 0 with properties:
pciBusID: 0000:02:00.0 name: GeForce GTX 1070 Ti computeCapability: 6.1
coreClock: 1.683GHz coreCount: 19 deviceMemorySize: 8.00GiB deviceMemoryBandwidth: 238.66GiB/s
2020-06-22 20:24:33.697234: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2020-06-22 20:24:33.747540: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2020-06-22 20:24:33.787573: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2020-06-22 20:24:33.810063: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2020-06-22 20:24:33.841474: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
2020-06-22 20:24:33.862787: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2020-06-22 20:24:33.907318: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2020-06-22 20:24:33.913612: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1703] Adding visible gpu devices: 0
2020-06-22 20:24:33.918093: I tensorflow/core/platform/cpu_feature_guard.cc:143] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2020-06-22 20:24:33.932784: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x2382acc1c40 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-06-22 20:24:33.939473: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
2020-06-22 20:24:33.944570: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1561] Found device 0 with properties:
pciBusID: 0000:02:00.0 name: GeForce GTX 1070 Ti computeCapability: 6.1
coreClock: 1.683GHz coreCount: 19 deviceMemorySize: 8.00GiB deviceMemoryBandwidth: 238.66GiB/s
2020-06-22 20:24:33.953910: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
2020-06-22 20:24:33.958772: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll
2020-06-22 20:24:33.963656: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cufft64_10.dll
2020-06-22 20:24:33.968210: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library curand64_10.dll
2020-06-22 20:24:33.973389: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusolver64_10.dll
2020-06-22 20:24:33.978058: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cusparse64_10.dll
2020-06-22 20:24:33.983547: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll
2020-06-22 20:24:33.990380: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1703] Adding visible gpu devices: 0
2020-06-22 20:24:35.338596: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-06-22 20:24:35.344643: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1108]      0
2020-06-22 20:24:35.348795: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1121] 0:   N
2020-06-22 20:24:35.353853: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1247] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 6284 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1070 Ti, pci bus id: 0000:02:00.0, compute capability: 6.1)
2020-06-22 20:24:35.369758: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x2384aa9f820 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2020-06-22 20:24:35.376320: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): GeForce GTX 1070 Ti, Compute Capability 6.1
tf.Tensor(122.478485, shape=(), dtype=float32)

When I tried to install the Object Detection API (last step), I can installed it successfully, but I cannot use GPU anymore (print-out show as case 1 I write above).

"Install the Object Detection API" step tell me to install packages from command below

python -m pip install .

setup.py file

"""Setup script for object_detection with TF2.0."""
import os
from setuptools import find_packages
from setuptools import setup

# Note: adding apache-beam to required packages causes conflict with
# tf-models-offical requirements. These packages request for incompatible
# oauth2client package.
REQUIRED_PACKAGES = [
    # Required for apache-beam with PY3
    'avro-python3',
    'apache-beam',
    'pillow',
    'lxml',
    'matplotlib',
    'Cython',
    'contextlib2',
    'tf-slim',
    'six',
    'pycocotools',
    'lvis',
    'scipy',
    'pandas',
    'tf-models-official'
]

setup(
    name='object_detection',
    version='0.1',
    install_requires=REQUIRED_PACKAGES,
    include_package_data=True,
    packages=(
        [p for p in find_packages() if p.startswith('object_detection')] +
        find_packages(where=os.path.join('.', 'slim'))),
    package_dir={
        'datasets': os.path.join('slim', 'datasets'),
        'nets': os.path.join('slim', 'nets'),
        'preprocessing': os.path.join('slim', 'preprocessing'),
        'deployment': os.path.join('slim', 'deployment'),
        'scripts': os.path.join('slim', 'scripts'),
    },
    description='Tensorflow Object Detection Library',
    python_requires='>3.6',
)

I notice that version of tensorflow was changed (from 2.2.0 to 2.4.0) after I installed the Object Detection API. Plus, my newly created conda env which installed tensorflow 2.2.0 could use GPU normally. Does the version of tensorflow make me cannot use GPU? If so, how do I install packages without changing the version of the other packages .

Answer 1

Tensorflow 2.2 uses CUDA 10.1 but after running the object-detection api your tensorflow is getting updated to 2.4, at which point GPU no longer is used to run the default test program.

This is because, Tensorflow 2.4.0 uses CUDA 11.0 . If you simply update cuda to the right version, you will have this working.