我使用的是配备 GPU 0：英特尔® 核芯显卡 630 和 GPU 1：GTX1050 TI 的笔记本电脑。我刚刚使用以下教程在 Anaconda 环境中完成了 YOLO 环境的设置 https://appliedmachinelearning.blog/2018/05/27/running-yolo-v2-for-real-time-object-detection-on-videos-images-via-darkflow/
问题是：每当我尝试使用 GPU 在 Anaconda 环境中使用 YOLO 渲染视频时
python flow --model cfg/yolo.cfg --load bin/yolov2.weights --demo videofile.mp4 --saveVideo --gpu 0.5
视频确实成功渲染，但是，我的 CPU 使用率几乎上升到 100%(任务管理器(，而我的 GPU 根本没有使用。我试图通过在末尾添加 --gpuName/gpu：1 来指定 GPU 名称，但仍然使用 CPU。以下是从 Anaconda 提示符复制的输出行。
(df) C:UsersUserVideosPC-programmingdarkflow-master>python flow --model cfg/yolo.cfg --load bin/yolov2.weights --demo videofile.mp4 --saveVideo --gpu 0.5
Parsing ./cfg/yolov2.cfg
Parsing cfg/yolo.cfg
Loading bin/yolov2.weights ...
Successfully identified 203934260 bytes
Finished in 0.022666454315185547s
Model has a coco model name, loading coco labels.
Building net ...
Source | Train? | Layer description | Output size
-------+--------+----------------------------------+---------------
| | input | (?, 608, 608, 3)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 608, 608, 32)
Load | Yep! | maxp 2x2p0_2 | (?, 304, 304, 32)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 304, 304, 64)
Load | Yep! | maxp 2x2p0_2 | (?, 152, 152, 64)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 152, 152, 128)
Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 152, 152, 64)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 152, 152, 128)
Load | Yep! | maxp 2x2p0_2 | (?, 76, 76, 128)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 76, 76, 256)
Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 76, 76, 128)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 76, 76, 256)
Load | Yep! | maxp 2x2p0_2 | (?, 38, 38, 256)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 38, 38, 512)
Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 38, 38, 256)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 38, 38, 512)
Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 38, 38, 256)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 38, 38, 512)
Load | Yep! | maxp 2x2p0_2 | (?, 19, 19, 512)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)
Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 19, 19, 512)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)
Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 19, 19, 512)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)
Load | Yep! | concat [16] | (?, 38, 38, 512)
Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 38, 38, 64)
Load | Yep! | local flatten 2x2 | (?, 19, 19, 256)
Load | Yep! | concat [27, 24] | (?, 19, 19, 1280)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)
Load | Yep! | conv 1x1p0_1 linear | (?, 19, 19, 425)
-------+--------+----------------------------------+---------------
GPU mode with 0.5 usage
2018-10-16 17:21:18.897583: W C:tf_jenkinshomeworkspacerel-winMwindowsPY36tensorflowcoreplatformcpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
2018-10-16 17:21:18.904824: W C:tf_jenkinshomeworkspacerel-winMwindowsPY36tensorflowcoreplatformcpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
Finished in 4.7458178997039795s
Press [ESC] to quit demo
0.719 FPS ......
然后，如果我尝试渲染图像，任务管理器仍然显示根本没有使用 GPU。
(df) C:UsersUserVideosPC-programmingdarkflow-master>python flow --model cfg/yolo.cfg --load bin/yolov2.weights --imgdir sample_img --gpu 0.9
Parsing ./cfg/yolov2.cfg
Parsing cfg/yolo.cfg
Loading bin/yolov2.weights ...
Successfully identified 203934260 bytes
Finished in 0.021943330764770508s
Model has a coco model name, loading coco labels.
Building net ...
Source | Train? | Layer description | Output size
-------+--------+----------------------------------+---------------
| | input | (?, 608, 608, 3)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 608, 608, 32)
Load | Yep! | maxp 2x2p0_2 | (?, 304, 304, 32)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 304, 304, 64)
Load | Yep! | maxp 2x2p0_2 | (?, 152, 152, 64)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 152, 152, 128)
Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 152, 152, 64)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 152, 152, 128)
Load | Yep! | maxp 2x2p0_2 | (?, 76, 76, 128)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 76, 76, 256)
Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 76, 76, 128)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 76, 76, 256)
Load | Yep! | maxp 2x2p0_2 | (?, 38, 38, 256)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 38, 38, 512)
Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 38, 38, 256)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 38, 38, 512)
Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 38, 38, 256)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 38, 38, 512)
Load | Yep! | maxp 2x2p0_2 | (?, 19, 19, 512)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)
Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 19, 19, 512)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)
Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 19, 19, 512)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)
Load | Yep! | concat [16] | (?, 38, 38, 512)
Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 38, 38, 64)
Load | Yep! | local flatten 2x2 | (?, 19, 19, 256)
Load | Yep! | concat [27, 24] | (?, 19, 19, 1280)
Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)
Load | Yep! | conv 1x1p0_1 linear | (?, 19, 19, 425)
-------+--------+----------------------------------+---------------
GPU mode with 0.9 usage
2018-10-16 17:07:30.439641: W C:tf_jenkinshomeworkspacerel-winMwindowsPY36tensorflowcoreplatformcpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
2018-10-16 17:07:30.449381: W C:tf_jenkinshomeworkspacerel-winMwindowsPY36tensorflowcoreplatformcpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
Finished in 5.261923789978027s
Forwarding 8 inputs ...
Total time = 10.975605964660645s / 8 inps = 0.7288891406778334 ips
Post processing 8 inputs ...
Total time = 0.48075294494628906s / 8 inps = 16.640563690969756 ips
怎么了>_
您是否在 Makefile 中做了一些更改，请在 Makefile 中启用 GPU 和 CUDNN。小心路径。
克隆 https://github.com/thtrieu/darkflow，
从 https://pjreddie.com/darknet/yolo/下载必要的.cfg和权重，
分别保存在Darkflow-master中的cfg文件夹和新的bin文件夹中。
conda create -n darkflow-env python=3.6
activate darkflow-env
pip install tensorflow-gpu(点，不是康达。此步骤还应自动安装 CUDA 和 cuDNN，无需单独下载。
conda install cython numpy
conda config --add channels conda-forge
conda install opencv
转到您的 Darkflow-master 文件夹并复制路径
cd 到路径(仍在使用 Anaconda 提示符(
python setup.py build_ext --inplace
python flow --model cfg/yolo.cfg --load bin/yolov2.weights --demo videofile.mp4 --saveVideo --gpu 0.7(视频文件.mp4是要渲染的视频，我直接把它放在 darkflow-master 文件夹中(
等待，您将在 Darkflow-master 文件夹中看到输出视频(在我配备 GTX1050 TI 显卡的笔记本电脑上，渲染速度约为 8.5 FPS(。
如果您遇到任何问题，例如需要Microsoft可视化C++构建工具，只需从Microsoft网站下载，请记住在安装的同时安装 SDK。有关此问题的详细信息，请参阅 https://github.com/thtrieu/darkflow/issues/788。