はじめに

　BiSeNetとLEDNetもinferenceはできたのですが、Trainingがちょっとあって停滞していて、先にこの、"Training-Time-Friendly Network for Real-Time Object Detection"の記事を書こうと思います。CenterNetやCornerNet-LiteはInferenceは速いのですが、Trainingははっきり言ってしんどいので、よほど強力なGPUがないとやってられません。
　その点、これは非常に短い時間でそれなりのTrainingができるので、独自のデータを学習させたり、COCOのカテゴリを減らしたらmAPがどう変わるかなどといったTrainingにかかわるいろいろなことが個人所有のPCやGPUで出来そうなので、有意義だと思いました。

0.注意

ディレクトリ名が途中でttfnet2,mmdetection2となったり、仮想環境がopen-mmlab2となるのは、いったん作った環境ではなく、再度この記事のために作ったためにこうなっています。皆さんは*2ではなく、2がついていないと思ってやってください。

1.準備

　INSTALL.mdの通りなのですが、Windows10なので適当に躓きます。

(base) F:\Users\sounansu\Anaconda3>conda create -n open-mmlab python=3.7 -y
Collecting package metadata (current_repodata.json): done
Solving environment: done

## Package Plan ##

  environment location: F:\Users\sounansu\Anaconda3\envs\open-mmlab

  added / updated specs:
    - python=3.7


The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    ca-certificates-2019.8.28  |                0         165 KB
    certifi-2019.9.11          |           py37_0         155 KB
    pip-19.2.2                 |           py37_0         1.7 MB
    python-3.7.4               |       h5263a28_0        14.7 MB
    setuptools-41.2.0          |           py37_0         665 KB
    sqlite-3.29.0              |       he774522_0         624 KB
    vc-14.1                    |       h0510ff6_4           6 KB
    wheel-0.33.6               |           py37_0          58 KB
    wincertstore-0.2           |           py37_0          14 KB
    ------------------------------------------------------------
                                           Total:        18.0 MB

The following NEW packages will be INSTALLED:

  ca-certificates    pkgs/main/win-64::ca-certificates-2019.8.28-0
  certifi            pkgs/main/win-64::certifi-2019.9.11-py37_0
  openssl            pkgs/main/win-64::openssl-1.1.1d-he774522_0
  pip                pkgs/main/win-64::pip-19.2.2-py37_0
  python             pkgs/main/win-64::python-3.7.4-h5263a28_0
  setuptools         pkgs/main/win-64::setuptools-41.2.0-py37_0
  sqlite             pkgs/main/win-64::sqlite-3.29.0-he774522_0
  vc                 pkgs/main/win-64::vc-14.1-h0510ff6_4
  vs2015_runtime     pkgs/main/win-64::vs2015_runtime-14.16.27012-hf0eaf9b_0
  wheel              pkgs/main/win-64::wheel-0.33.6-py37_0
  wincertstore       pkgs/main/win-64::wincertstore-0.2-py37_0



Downloading and Extracting Packages
python-3.7.4         | 14.7 MB   | ################################################################################################################## | 100%
certifi-2019.9.11    | 155 KB    | ################################################################################################################## | 100%
ca-certificates-2019 | 165 KB    | ################################################################################################################## | 100%
sqlite-3.29.0        | 624 KB    | ################################################################################################################## | 100%
wincertstore-0.2     | 14 KB     | ################################################################################################################## | 100%
setuptools-41.2.0    | 665 KB    | ################################################################################################################## | 100%
wheel-0.33.6         | 58 KB     | ################################################################################################################## | 100%
vc-14.1              | 6 KB      | ################################################################################################################## | 100%
pip-19.2.2           | 1.7 MB    | ################################################################################################################## | 100%
Preparing transaction: done
Verifying transaction: done
Executing transaction: done
#
# To activate this environment, use
#
#     $ conda activate open-mmlab
#
# To deactivate an active environment, use
#
#     $ conda deactivate

(base) F:\Users\sounansu\Anaconda3>conda activate open-mmlab

(open-mmlab) F:\Users\sounansu\Anaconda3>conda install cython
Collecting package metadata (current_repodata.json): done
Solving environment: done

## Package Plan ##

  environment location: F:\Users\sounansu\Anaconda3\envs\open-mmlab

  added / updated specs:
    - cython


The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    cython-0.29.13             |   py37ha925a31_0         1.8 MB
    ------------------------------------------------------------
                                           Total:         1.8 MB

The following NEW packages will be INSTALLED:

  cython             pkgs/main/win-64::cython-0.29.13-py37ha925a31_0


Proceed ([y]/n)? y


Downloading and Extracting Packages
cython-0.29.13       | 1.8 MB    | ################################################################################################################## | 100%
Preparing transaction: done
Verifying transaction: done
Executing transaction: done

(open-mmlab) F:\Users\sounansu\Anaconda3>conda install pytorch torchvision cudatoolkit=10.0 -c pytorch
Collecting package metadata (current_repodata.json): done
Solving environment: done

## Package Plan ##

  environment location: F:\Users\sounansu\Anaconda3\envs\open-mmlab

  added / updated specs:
    - cudatoolkit=10.0
    - pytorch
    - torchvision


The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    blas-1.0                   |              mkl           6 KB
    mkl_fft-1.0.14             |   py37h14836fe_0         118 KB
    mkl_random-1.0.2           |   py37h343c172_0         272 KB
    ninja-1.9.0                |   py37h74a9793_0         238 KB
    numpy-1.16.5               |   py37h19fb1c0_0          50 KB
    olefile-0.46               |           py37_0          49 KB
    pillow-6.1.0               |   py37hdc69c19_0         636 KB
    pycparser-2.19             |           py37_0         170 KB
    pytorch-1.2.0              |py3.7_cuda100_cudnn7_1       479.3 MB  pytorch
    six-1.12.0                 |           py37_0          23 KB
    torchvision-0.4.0          |       py37_cu100         2.3 MB  pytorch
    ------------------------------------------------------------
                                           Total:       483.1 MB

The following NEW packages will be INSTALLED:

  blas               pkgs/main/win-64::blas-1.0-mkl
  cffi               pkgs/main/win-64::cffi-1.12.3-py37h7a1dbc1_0
  cudatoolkit        pkgs/main/win-64::cudatoolkit-10.0.130-0
  freetype           pkgs/main/win-64::freetype-2.9.1-ha9979f8_1
  icc_rt             pkgs/main/win-64::icc_rt-2019.0.0-h0cc432a_1
  intel-openmp       pkgs/main/win-64::intel-openmp-2019.4-245
  jpeg               pkgs/main/win-64::jpeg-9b-hb83a4c4_2
  libpng             pkgs/main/win-64::libpng-1.6.37-h2a8f88b_0
  libtiff            pkgs/main/win-64::libtiff-4.0.10-hb898794_2
  mkl                pkgs/main/win-64::mkl-2019.4-245
  mkl-service        pkgs/main/win-64::mkl-service-2.3.0-py37hb782905_0
  mkl_fft            pkgs/main/win-64::mkl_fft-1.0.14-py37h14836fe_0
  mkl_random         pkgs/main/win-64::mkl_random-1.0.2-py37h343c172_0
  ninja              pkgs/main/win-64::ninja-1.9.0-py37h74a9793_0
  numpy              pkgs/main/win-64::numpy-1.16.5-py37h19fb1c0_0
  numpy-base         pkgs/main/win-64::numpy-base-1.16.5-py37hc3f5095_0
  olefile            pkgs/main/win-64::olefile-0.46-py37_0
  pillow             pkgs/main/win-64::pillow-6.1.0-py37hdc69c19_0
  pycparser          pkgs/main/win-64::pycparser-2.19-py37_0
  pytorch            pytorch/win-64::pytorch-1.2.0-py3.7_cuda100_cudnn7_1
  six                pkgs/main/win-64::six-1.12.0-py37_0
  tk                 pkgs/main/win-64::tk-8.6.8-hfa6e2cd_0
  torchvision        pytorch/win-64::torchvision-0.4.0-py37_cu100
  xz                 pkgs/main/win-64::xz-5.2.4-h2fa13f4_4
  zlib               pkgs/main/win-64::zlib-1.2.11-h62dcd97_3
  zstd               pkgs/main/win-64::zstd-1.3.7-h508b16e_0


Proceed ([y]/n)? y


Downloading and Extracting Packages
mkl_fft-1.0.14       | 118 KB    | ################################################################################################################## | 100%
olefile-0.46         | 49 KB     | ################################################################################################################## | 100%
pycparser-2.19       | 170 KB    | ################################################################################################################## | 100%
mkl_random-1.0.2     | 272 KB    | ################################################################################################################## | 100%
six-1.12.0           | 23 KB     | ################################################################################################################## | 100%
ninja-1.9.0          | 238 KB    | ################################################################################################################## | 100%
numpy-1.16.5         | 50 KB     | ################################################################################################################## | 100%
torchvision-0.4.0    | 2.3 MB    | ################################################################################################################## | 100%
pillow-6.1.0         | 636 KB    | ################################################################################################################## | 100%
blas-1.0             | 6 KB      | ################################################################################################################## | 100%
pytorch-1.2.0        | 479.3 MB  | ################################################################################################################## | 100%
Preparing transaction: done
Verifying transaction: done
Executing transaction: done

次にリポジトリを取ってきましょう。

(open-mmlab) F:\Users\sounansu\Anaconda3>git clone https://github.com/ZJULearning/ttfnet.git
'git' は、内部コマンドまたは外部コマンド、
操作可能なプログラムまたはバッチ ファイルとして認識されていません。

(open-mmlab) F:\Users\sounansu\Anaconda3>

gitはデフォルトではないので先に入れて、

(open-mmlab) F:\Users\sounansu\Anaconda3>conda install git
Collecting package metadata (current_repodata.json): done
Solving environment: done

## Package Plan ##

  environment location: F:\Users\sounansu\Anaconda3\envs\open-mmlab

  added / updated specs:
    - git


The following NEW packages will be INSTALLED:

  git                pkgs/main/win-64::git-2.20.1-h6bb4b03_0


Proceed ([y]/n)? y

Preparing transaction: done
Verifying transaction: done
Executing transaction: done

(open-mmlab) F:\Users\sounansu\Anaconda3>git clone https://github.com/ZJULearning/ttfnet.git
Cloning into 'ttfnet'...
remote: Enumerating objects: 4913, done.
remote: Counting objects: 100% (4913/4913), done.
remote: Compressing objects: 100% (1647/1647), done.
remote: Total 4913 (delta 3279), reused 4873 (delta 3240), pack-reused 0
Receiving objects: 100% (4913/4913), 5.45 MiB | 1.59 MiB/s, done.
Resolving deltas: 100% (3279/3279), done.

下準備をするのですが、、、

(open-mmlab) F:\Users\sounansu\Anaconda3>cd ttfnet
(open-mmlab) F:\Users\sounansu\Anaconda3\ttfnet>python setup.py develop

ものすごく時間がかかって

C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Tools\MSVC\14.16.27023\bin\HostX86\x64\link.exe /nologo /INCREMENTAL:NO /LTCG /DLL /MANIFEST:EMBED,ID=2 /MANIFESTUAC:NO /LIBPATH:F:\Users\sounansu\Anaconda3\envs\open-mmlab\lib\site-packages\torch\lib "/LIBPATH:C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\lib/x64" /LIBPATH:F:\Users\sounansu\Anaconda3\envs\open-mmlab\libs /LIBPATH:F:\Users\sounansu\Anaconda3\envs\open-mmlab\PCbuild\amd64 "/LIBPATH:C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Tools\MSVC\14.16.27023\ATLMFC\lib\x64" "/LIBPATH:C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Tools\MSVC\14.16.27023\lib\x64" "/LIBPATH:C:\Program Files (x86)\Windows Kits\NETFXSDK\4.6.1\lib\um\x64" "/LIBPATH:C:\Program Files (x86)\Windows Kits\10\lib\10.0.17763.0\ucrt\x64" "/LIBPATH:C:\Program Files (x86)\Windows Kits\10\lib\10.0.17763.0\um\x64" cudart.lib c10.lib c10_cuda.lib torch.lib torch_python.lib _C.lib /EXPORT:PyInit_sigmoid_focal_loss_cuda build\temp.win-amd64-3.7\Release\mmdet\ops\sigmoid_focal_loss\src/sigmoid_focal_loss.obj build\temp.win-amd64-3.7\Release\mmdet\ops\sigmoid_focal_loss\src/sigmoid_focal_loss_cuda.obj /OUT:build\lib.win-amd64-3.7\mmdet\ops\sigmoid_focal_loss\sigmoid_focal_loss_cuda.cp37-win_amd64.pyd /IMPLIB:build\temp.win-amd64-3.7\Release\mmdet\ops\sigmoid_focal_loss\src\sigmoid_focal_loss_cuda.cp37-win_amd64.lib
   ライブラリ build\temp.win-amd64-3.7\Release\mmdet\ops\sigmoid_focal_loss\src\sigmoid_focal_loss_cuda.cp37-win_amd64.lib とオブジェクト build\temp.win-amd64-3.7\Release\mmdet\ops\sigmoid_focal_loss\src\sigmoid_focal_loss_cuda.cp37-win_amd64.exp を作成中
sigmoid_focal_loss_cuda.obj : error LNK2001: 外部シンボル ""public: long * __cdecl at::Tensor::data<long>(void)const " (??$data@J@Tensor@at@@QEBAPEAJXZ)" は未解決です。
build\lib.win-amd64-3.7\mmdet\ops\sigmoid_focal_loss\sigmoid_focal_loss_cuda.cp37-win_amd64.pyd : fatal error LNK1120: 1 件の未解決の外部参照
error: command 'C:\\Program Files (x86)\\Microsoft Visual Studio\\2017\\Community\\VC\\Tools\\MSVC\\14.16.27023\\bin\\HostX86\\x64\\link.exe' failed with exit status 1120

(open-mmlab) F:\Users\sounansu\Anaconda3\ttfnet>

あらあらまあまあエラーが出てしまっています。ttfnetのissueをみてもそれらしいのがないので、ttfnetがつかっている環境であるopen-mmlabの方にヒントがないか見に行きましょう。

2.MMDetection open-mmlab

open-mmlabは、MMDetection: Open MMLab Detection Toolbox and Benchmarkであるように、COCOのチャレンジの際にMMDetチームが使ったコードをもとに整備されているもので、open-mmlabにはmmcv, mmskelton, mmsr, mmaction, mmstyle等のコードが入っています。
　MMDetectionには、RPN, Fast R-CNN, Faster R-CNN, Mask R-CNN, Cascade R-CNN, Cascade Mask R-CNN, SSD, RetinaNet, GHM, Mask Scoring R-CNN, FCOS, Double-Head R-CNN, Grid R-CNN (Plus), Hybrid Task Cascade, Libra R-CNN, Guided Anchoringのベンチマークとモデルズ―があり、ほぼ同じ環境での構造の比較や、すでにトレーニングされたウエイトデータもあるので、インファレンスをお試し等もできます。
　よくある、githubだと、トレーニングしかできなかったり、インファレンスしかできなかったり、トレーニングデータがこっちはCOCO、こっちはCityScape等で比較できないのですが、そこも含めて比べられるので、一度覗いてみてはいかがでしょうか？
　と、今回はTTFNetを実行するために、mmdetectionをまずは動かすということで。

2.1 Install

Install mmdetectionの初めの方はTTFNetと同じなので、mmdetectionのgitを取ってくるところから、

(open-mmlab2) F:\Users\sounansu\Anaconda3>git clone https://github.com/open-mmlab/mmdetection.git mmdetection2
Cloning into 'mmdetection2'...
remote: Enumerating objects: 5, done.
remote: Counting objects: 100% (5/5), done.
remote: Compressing objects: 100% (5/5), done.
remote: Total 5254 (delta 0), reused 0 (delta 0), pack-reused 5249
Receiving objects: 100% (5254/5254), 5.01 MiB | 1.25 MiB/s, done.
Resolving deltas: 100% (3528/3528), done.

(open-mmlab2) F:\Users\sounansu\Anaconda3>cd mmdetection2

(open-mmlab2) F:\Users\sounansu\Anaconda3\mmdetection2>set PYTHONPATH=f:\Users\sounansu\Anaconda3\mmdetection2

(open-mmlab2) F:\Users\sounansu\Anaconda3\mmdetection2>python setup.py develop
いろいろあって
cl : コマンド ライン error D8021 : 数値型引数 '/Wno-cpp' は無効です。
error: Setup script exited with error: command 'C:\\Program Files (x86)\\Microsoft Visual Studio\\2017\\Community\\VC\\Tools\\MSVC\\14.16.27023\\bin\\HostX86\\x64\\cl.exe' failed with exit status 2

issueをみると
issue1088に
1.Install community version of pycocotools for windows (official repository does not support windows)
とあるのでそれを入れてみると、

(open-mmlab2) F:\Users\sounansu\Anaconda3\mmdetection2> pip install git+https://github.com/philferriere/cocoapi.git#subdirectory=PythonAPI
Collecting git+https://github.com/philferriere/cocoapi.git#subdirectory=PythonAPI
  Cloning https://github.com/philferriere/cocoapi.git to e:\temp\pip-req-build-n5g737fa
  Running command git clone -q https://github.com/philferriere/cocoapi.git 'E:\TEMP\pip-req-build-n5g737fa'
Building wheels for collected packages: pycocotools
  Building wheel for pycocotools (setup.py) ... done
  Created wheel for pycocotools: filename=pycocotools-2.0-cp37-cp37m-win_amd64.whl size=82677 sha256=16c7e7544004c242838d2be3c711f3c2a8a12c7f235bcf342401aad4275812f1
  Stored in directory: E:\TEMP\pip-ephem-wheel-cache-i9qk_qxe\wheels\69\2b\12\2fa959e49f73d26cff202c2f4e5079096c9c57c8a8509fd75c
Successfully built pycocotools
ERROR: mmdet 1.0rc0+d88ad8f requires matplotlib, which is not installed.
ERROR: mmdet 1.0rc0+d88ad8f requires mmcv>=0.2.10, which is not installed.
ERROR: mmdet 1.0rc0+d88ad8f requires terminaltables, which is not installed.
Installing collected packages: pycocotools
Successfully installed pycocotools-2.0

と文句を言われるので、この3つを入れましょう。
まず、matplotlibはcondaで

(open-mmlab2) F:\Users\sounansu\Anaconda3\mmdetection2>conda install matplotlib
（ログが消えちゃうんですよね。。）

mmcvはInstallにあるように

(open-mmlab2) F:\Users\sounansu\Anaconda3>git clone https://github.com/open-mmlab/mmcv.git
(open-mmlab2) F:\Users\sounansu\Anaconda3>cd mmcv
(open-mmlab2) F:\Users\sounansu\Anaconda3\mmcv>pip install -e .
Obtaining file:///F:/Users/sounansu/Anaconda3/mmcv
Requirement already satisfied: numpy>=1.11.1 in f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages (from mmcv==0.2.13) (1.16.5)
Requirement already satisfied: pyyaml in f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages (from mmcv==0.2.13) (5.1.2)
Requirement already satisfied: six in f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages (from mmcv==0.2.13) (1.12.0)
Requirement already satisfied: addict in f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages (from mmcv==0.2.13) (2.2.1)
Requirement already satisfied: requests in f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages (from mmcv==0.2.13) (2.22.0)
Requirement already satisfied: opencv-python in f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages (from mmcv==0.2.13) (4.1.1.26)
Requirement already satisfied: Cython in f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages (from mmcv==0.2.13) (0.29.13)
Requirement already satisfied: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages (from requests->mmcv==0.2.13) (1.25.6)
Requirement already satisfied: chardet<3.1.0,>=3.0.2 in f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages (from requests->mmcv==0.2.13) (3.0.4)
Requirement already satisfied: certifi>=2017.4.17 in f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages (from requests->mmcv==0.2.13) (2019.9.11)
Requirement already satisfied: idna<2.9,>=2.5 in f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages (from requests->mmcv==0.2.13) (2.8)
ERROR: mmdet 1.0rc0+d88ad8f requires terminaltables, which is not installed.
Installing collected packages: mmcv
  Found existing installation: mmcv 0.2.13
    Uninstalling mmcv-0.2.13:
      Successfully uninstalled mmcv-0.2.13
  Running setup.py develop for mmcv
Successfully installed mmcv

(open-mmlab2) F:\Users\sounansu\Anaconda3\mmcv>

あれ？すでに入れていたようですね。（なぜさっきは、入ってないと表示された？）
最後に、terminaltablesは

(open-mmlab2) F:\Users\sounansu\Anaconda3\mmdetection>pip install terminaltables
Collecting terminaltables
Installing collected packages: terminaltables
Successfully installed terminaltables-3.1.0

としていれて、改めてpython setup.py developをすると

(open-mmlab2) F:\Users\sounansu\Anaconda3\mmdetection2>python setup.py develop
F:\Users\sounansu\Anaconda3\envs\open-mmlab2\lib\site-packages\setuptools\dist.py:474: UserWarning: Normalizing '1.0.rc0+d88ad8f' to '1.0rc0+d88ad8f'
  normalized_version,
running develop
running egg_info
writing mmdet.egg-info\PKG-INFO
writing dependency_links to mmdet.egg-info\dependency_links.txt
writing requirements to mmdet.egg-info\requires.txt
writing top-level names to mmdet.egg-info\top_level.txt
reading manifest file 'mmdet.egg-info\SOURCES.txt'
writing manifest file 'mmdet.egg-info\SOURCES.txt'
running build_ext
F:\Users\sounansu\Anaconda3\envs\open-mmlab2\lib\site-packages\torch\utils\cpp_extension.py:189: UserWarning: Error checking compiler version for cl: 'utf-8' codec can't decode byte 0x8e in position 132: invalid start byte
  warnings.warn('Error checking compiler version for {}: {}'.format(compiler, error))
copying build\lib.win-amd64-3.7\mmdet\ops\nms\soft_nms_cpu.cp37-win_amd64.pyd -> mmdet\ops\nms
copying build\lib.win-amd64-3.7\mmdet\ops\nms\nms_cpu.cp37-win_amd64.pyd -> mmdet\ops\nms
copying build\lib.win-amd64-3.7\mmdet\ops\nms\nms_cuda.cp37-win_amd64.pyd -> mmdet\ops\nms
copying build\lib.win-amd64-3.7\mmdet\ops\roi_align\roi_align_cuda.cp37-win_amd64.pyd -> mmdet\ops\roi_align
copying build\lib.win-amd64-3.7\mmdet\ops\roi_pool\roi_pool_cuda.cp37-win_amd64.pyd -> mmdet\ops\roi_pool
copying build\lib.win-amd64-3.7\mmdet\ops\dcn\deform_conv_cuda.cp37-win_amd64.pyd -> mmdet\ops\dcn
copying build\lib.win-amd64-3.7\mmdet\ops\dcn\deform_pool_cuda.cp37-win_amd64.pyd -> mmdet\ops\dcn
copying build\lib.win-amd64-3.7\mmdet\ops\sigmoid_focal_loss\sigmoid_focal_loss_cuda.cp37-win_amd64.pyd -> mmdet\ops\sigmoid_focal_loss
copying build\lib.win-amd64-3.7\mmdet\ops\masked_conv\masked_conv2d_cuda.cp37-win_amd64.pyd -> mmdet\ops\masked_conv
Creating f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\mmdet.egg-link (link to .)
Adding mmdet 1.0rc0+d88ad8f to easy-install.pth file

Installed f:\users\sounansu\anaconda3\mmdetection2
Processing dependencies for mmdet==1.0rc0+d88ad8f
Searching for opencv-python-headless>=4.1.1
Reading https://pypi.org/simple/opencv-python-headless/
Downloading https://files.pythonhosted.org/packages/9a/a2/1a7feb6565cd932d9c9d06c9bde22fe9345a348ee3ed4f55f2457e95b111/opencv_python_headless-4.1.1.26-cp37-cp37m-win_amd64.whl#sha256=a6e9114c1e57280f47c796d87ffaea49aa3e3bcb4ed1ccde18c500b8cd143c73
Best match: opencv-python-headless 4.1.1.26
Processing opencv_python_headless-4.1.1.26-cp37-cp37m-win_amd64.whl
Installing opencv_python_headless-4.1.1.26-cp37-cp37m-win_amd64.whl to f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
writing requirements to f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\opencv_python_headless-4.1.1.26-py3.7-win-amd64.egg\EGG-INFO\requires.txt
Adding opencv-python-headless 4.1.1.26 to easy-install.pth file

Installed f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\opencv_python_headless-4.1.1.26-py3.7-win-amd64.egg
Searching for imgaug<0.2.7,>=0.2.5
Reading https://pypi.org/simple/imgaug/
Downloading https://files.pythonhosted.org/packages/ad/2e/748dbb7bb52ec8667098bae9b585f448569ae520031932687761165419a2/imgaug-0.2.6.tar.gz#sha256=82a5052d2daabc1b8de04b5fade44d59f3376c429880dd78e759028c67f3c8f3
Best match: imgaug 0.2.6
Processing imgaug-0.2.6.tar.gz
Writing E:\TEMP\easy_install-9y1j8wcb\imgaug-0.2.6\setup.cfg
Running imgaug-0.2.6\setup.py -q bdist_egg --dist-dir E:\TEMP\easy_install-9y1j8wcb\imgaug-0.2.6\egg-dist-tmp-ln0f2ba0
  File "build\bdist.win-amd64\egg\imgaug\bak\affine.py", line 108
    return shear = shear
                 ^
SyntaxError: invalid syntax

Sorry: IndentationError: unexpected indent (parameters-testcode.py, line 2)
zip_safe flag not set; analyzing archive contents...
imgaug.__pycache__.imgaug.cpython-37: module references __file__
creating f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\imgaug-0.2.6-py3.7.egg
Extracting imgaug-0.2.6-py3.7.egg to f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Sorry: IndentationError: unexpected indent (parameters-testcode.py, line 2)
  File "f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\imgaug-0.2.6-py3.7.egg\imgaug\bak\affine.py", line 108
    return shear = shear
                 ^
SyntaxError: invalid syntax

Adding imgaug 0.2.6 to easy-install.pth file

Installed f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\imgaug-0.2.6-py3.7.egg
Searching for scipy
Reading https://pypi.org/simple/scipy/
Downloading https://files.pythonhosted.org/packages/50/eb/defa40367863304e1ef01c6572584c411446a5f29bdd9dc90f91509e9144/scipy-1.3.1-cp37-cp37m-win_amd64.whl#sha256=a9d606d11eb2eec7ef893eb825017fbb6eef1e1d0b98a5b7fc11446ebeb2b9b1
Best match: scipy 1.3.1
Processing scipy-1.3.1-cp37-cp37m-win_amd64.whl
Installing scipy-1.3.1-cp37-cp37m-win_amd64.whl to f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
writing requirements to f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\scipy-1.3.1-py3.7-win-amd64.egg\EGG-INFO\requires.txt
Adding scipy 1.3.1 to easy-install.pth file

Installed f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\scipy-1.3.1-py3.7-win-amd64.egg
Searching for scikit-image>=0.15
Reading https://pypi.org/simple/scikit-image/
Downloading https://files.pythonhosted.org/packages/79/16/c5a36a03f90d4a246791d4ff1879f1868e1c5db58fac9f03427395c5e2d6/scikit_image-0.15.0-cp37-none-win_amd64.whl#sha256=f0491621a6b1d2828d47acaf41d3167a647eaae44ef8fcf83b72eb3e1cc7ac51
Best match: scikit-image 0.15.0
Processing scikit_image-0.15.0-cp37-none-win_amd64.whl
Installing scikit_image-0.15.0-cp37-none-win_amd64.whl to f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
writing requirements to f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\scikit_image-0.15.0-py3.7-win-amd64.egg\EGG-INFO\requires.txt
Adding scikit-image 0.15.0 to easy-install.pth file
Installing skivi-script.py script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts
Installing skivi.exe script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts

Installed f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\scikit_image-0.15.0-py3.7-win-amd64.egg
Searching for networkx>=2.0
Reading https://pypi.org/simple/networkx/
Downloading https://files.pythonhosted.org/packages/85/08/f20aef11d4c343b557e5de6b9548761811eb16e438cee3d32b1c66c8566b/networkx-2.3.zip#sha256=8311ddef63cf5c5c5e7c1d0212dd141d9a1fe3f474915281b73597ed5f1d4e3d
Best match: networkx 2.3
Processing networkx-2.3.zip
Writing E:\TEMP\easy_install-0_rr4pbl\networkx-2.3\setup.cfg
Running networkx-2.3\setup.py -q bdist_egg --dist-dir E:\TEMP\easy_install-0_rr4pbl\networkx-2.3\egg-dist-tmp-8bjpvqs3
warning: no files found matching '*.html' under directory 'doc'
warning: no files found matching '*.css' under directory 'doc'
warning: no previously-included files matching '*~' found anywhere in distribution
warning: no previously-included files matching '*.pyc' found anywhere in distribution
warning: no previously-included files matching '.svn' found anywhere in distribution
no previously-included directories found matching 'doc\build'
no previously-included directories found matching 'doc\auto_examples'
no previously-included directories found matching 'doc\modules'
no previously-included directories found matching 'doc\reference\generated'
no previously-included directories found matching 'doc\reference\algorithms\generated'
no previously-included directories found matching 'doc\reference\classes\generated'
no previously-included directories found matching 'doc\reference\readwrite\generated'
creating f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\networkx-2.3-py3.7.egg
Extracting networkx-2.3-py3.7.egg to f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Adding networkx 2.3 to easy-install.pth file

Installed f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\networkx-2.3-py3.7.egg
Searching for imageio>=2.0.1
Reading https://pypi.org/simple/imageio/
Downloading https://files.pythonhosted.org/packages/af/0a/943c965d372dae0b1f1482677d29030ab834351a61a9a632fd62f27f1523/imageio-2.5.0-py3-none-any.whl#sha256=1a2bbbb7cd38161340fa3b14d806dfbf914abf3ee6fd4592af2afb87d049f209
Best match: imageio 2.5.0
Processing imageio-2.5.0-py3-none-any.whl
Installing imageio-2.5.0-py3-none-any.whl to f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
writing requirements to f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\imageio-2.5.0-py3.7.egg\EGG-INFO\requires.txt
Adding imageio 2.5.0 to easy-install.pth file
Installing imageio_download_bin-script.py script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts
Installing imageio_download_bin.exe script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts
Installing imageio_remove_bin-script.py script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts
Installing imageio_remove_bin.exe script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts

Installed f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\imageio-2.5.0-py3.7.egg
Searching for PyWavelets>=0.4.0
Reading https://pypi.org/simple/PyWavelets/
Downloading https://files.pythonhosted.org/packages/71/57/1b6c06f2b681975e5b43ff13ba127031f668d65b592cc56fae6445b675c7/PyWavelets-1.0.3-cp37-none-win_amd64.whl#sha256=ab02363467ee3cb222c5b425bc53453270ddae72bce313e72fd14616692d725a
Best match: PyWavelets 1.0.3
Processing PyWavelets-1.0.3-cp37-none-win_amd64.whl
Installing PyWavelets-1.0.3-cp37-none-win_amd64.whl to f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
writing requirements to f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\PyWavelets-1.0.3-py3.7-win-amd64.egg\EGG-INFO\requires.txt
Adding PyWavelets 1.0.3 to easy-install.pth file

Installed f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\pywavelets-1.0.3-py3.7-win-amd64.egg
Searching for decorator>=4.3.0
Reading https://pypi.org/simple/decorator/
Downloading https://files.pythonhosted.org/packages/5f/88/0075e461560a1e750a0dcbf77f1d9de775028c37a19a346a6c565a257399/decorator-4.4.0-py2.py3-none-any.whl#sha256=f069f3a01830ca754ba5258fde2278454a0b5b79e0d7f5c13b3b97e57d4acff6
Best match: decorator 4.4.0
Processing decorator-4.4.0-py2.py3-none-any.whl
Installing decorator-4.4.0-py2.py3-none-any.whl to f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Adding decorator 4.4.0 to easy-install.pth file

Installed f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\decorator-4.4.0-py3.7.egg
Searching for albumentations==0.3.3
Best match: albumentations 0.3.3
Processing albumentations-0.3.3-py3.7.egg
albumentations 0.3.3 is already the active version in easy-install.pth

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\albumentations-0.3.3-py3.7.egg
Searching for imagecorruptions==1.0.0
Best match: imagecorruptions 1.0.0
Processing imagecorruptions-1.0.0-py3.7.egg
imagecorruptions 1.0.0 is already the active version in easy-install.pth

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\imagecorruptions-1.0.0-py3.7.egg
Searching for torchvision==0.4.0
Best match: torchvision 0.4.0
Adding torchvision 0.4.0 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for torch==1.2.0
Best match: torch 1.2.0
Adding torch 1.2.0 to easy-install.pth file
Installing convert-caffe2-to-onnx-script.py script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts
Installing convert-caffe2-to-onnx.exe script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts
Installing convert-onnx-to-caffe2-script.py script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts
Installing convert-onnx-to-caffe2.exe script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for pycocotools==2.0
Best match: pycocotools 2.0
Adding pycocotools 2.0 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for terminaltables==3.1.0
Best match: terminaltables 3.1.0
Adding terminaltables 3.1.0 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for six==1.12.0
Best match: six 1.12.0
Adding six 1.12.0 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for matplotlib==3.1.1
Best match: matplotlib 3.1.1
Adding matplotlib 3.1.1 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for numpy==1.16.5
Best match: numpy 1.16.5
Adding numpy 1.16.5 to easy-install.pth file
Installing f2py-script.py script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts
Installing f2py.exe script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for mmcv==0.2.13
Best match: mmcv 0.2.13
mmcv 0.2.13 is already the active version in easy-install.pth

Using f:\users\sounansu\anaconda3\mmcv
Searching for PyYAML==5.1.2
Best match: PyYAML 5.1.2
Adding PyYAML 5.1.2 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for opencv-python==4.1.1.26
Best match: opencv-python 4.1.1.26
Adding opencv-python 4.1.1.26 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for Pillow==6.1.0
Best match: Pillow 6.1.0
Adding Pillow 6.1.0 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for python-dateutil==2.8.0
Best match: python-dateutil 2.8.0
Adding python-dateutil 2.8.0 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for pyparsing==2.4.2
Best match: pyparsing 2.4.2
Adding pyparsing 2.4.2 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for kiwisolver==1.1.0
Best match: kiwisolver 1.1.0
Adding kiwisolver 1.1.0 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for cycler==0.10.0
Best match: cycler 0.10.0
Adding cycler 0.10.0 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for Cython==0.29.13
Best match: Cython 0.29.13
Adding Cython 0.29.13 to easy-install.pth file
Installing cygdb-script.py script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts
Installing cygdb.exe script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts
Installing cython-script.py script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts
Installing cython.exe script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts
Installing cythonize-script.py script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts
Installing cythonize.exe script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for requests==2.22.0
Best match: requests 2.22.0
Adding requests 2.22.0 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for addict==2.2.1
Best match: addict 2.2.1
Adding addict 2.2.1 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for setuptools==41.2.0
Best match: setuptools 41.2.0
Adding setuptools 41.2.0 to easy-install.pth file
Installing easy_install-script.py script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts
Installing easy_install.exe script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for idna==2.8
Best match: idna 2.8
Adding idna 2.8 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for certifi==2019.9.11
Best match: certifi 2019.9.11
Adding certifi 2019.9.11 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for urllib3==1.25.6
Best match: urllib3 1.25.6
Adding urllib3 1.25.6 to easy-install.pth file

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Searching for chardet==3.0.4
Best match: chardet 3.0.4
Adding chardet 3.0.4 to easy-install.pth file
Installing chardetect-script.py script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts
Installing chardetect.exe script to F:\Users\sounansu\Anaconda3\envs\open-mmlab2\Scripts

Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages
Finished processing dependencies for mmdet==1.0rc0+d88ad8f

(open-mmlab2) F:\Users\sounansu\Anaconda3\mmdetection2>

なんかエラーっぽいものが出ていますが、Finishしていますね。問題が出たらその時対応するということで。

2.2 モデルズーでテスト

　いきなり、TTFNetは上手くいかないかもしれないので（うまくいきませんでした。）まずはmmdetectionのご本尊のMODEL ZOOでテストしましょう。
　MODEL ZOOのmodelをブラウザでクリックするかして、ウエイトデータをダウンロードしてください。
　ダウンロードしたssd300_coco_vgg16_caffe_120e_20181221-84d7110b.pthをmmdetectionのフォルダにおいて、GETTING_STARTEDのpython tools/test.py configs/faster_rcnn_r50_fpn_1x.py checkpoints/faster_rcnn_r50_fpn_1x_20181010-3d1b3351.pth --showを参考にSSDを実行してみましょう。

(open-mmlab2) F:\Users\sounansu\Anaconda3\mmdetection2>python tools\test.py configs\ssd300_coco.py ssd300_coco_vgg16_caffe_120e_20181221-84d7110b.pth --show
loading annotations into memory...
Traceback (most recent call last):
  File "tools\test.py", line 224, in <module>
    main()
  File "tools\test.py", line 161, in main
    dataset = build_dataset(cfg.data.test)
  File "f:\Users\sounansu\Anaconda3\mmdetection2\mmdet\datasets\builder.py", line 39, in build_dataset
    dataset = build_from_cfg(cfg, DATASETS, default_args)
  File "f:\Users\sounansu\Anaconda3\mmdetection2\mmdet\utils\registry.py", line 76, in build_from_cfg
    return obj_cls(**args)
  File "f:\Users\sounansu\Anaconda3\mmdetection2\mmdet\datasets\custom.py", line 64, in __init__
    self.img_infos = self.load_annotations(self.ann_file)
  File "f:\Users\sounansu\Anaconda3\mmdetection2\mmdet\datasets\coco.py", line 27, in load_annotations
    self.coco = COCO(ann_file)
  File "F:\Users\sounansu\Anaconda3\envs\open-mmlab2\lib\site-packages\pycocotools\coco.py", line 85, in __init__
    dataset = json.load(open(annotation_file, 'r'))
FileNotFoundError: [Errno 2] No such file or directory: 'data/coco/annotations/instances_val2017.json'

data/coco/annotations以下がないといわれていますが、そもそもmmdetectionの下にdata自体がありません。INSTALLのprepare datasetsを参考にCOCOの部分を作ってください。VOCやCityScapeも、私は作りましたが。

(open-mmlab2) F:\Users\sounansu\Anaconda3\mmdetection2>python tools\test.py configs\ssd300_coco.py ssd300_coco_vgg16_caffe_120e_20181221-84d7110b.pth --show
loading annotations into memory...
Done (t=0.94s)
creating index...
index created!
[                                                  ] 0/5000, elapsed: 0s, ETA:

これがプロンプロに出て、

このような絵が出れば正解です。
お疲れさまでした。(TTFNetにはまだまだ。。。。)

3 TTFNetの続き

　せっかくmmdetectionができたので、TTFNetのconfigとウエイトデータを持って行って実行したのですが、TTFNetで使っている一部のネットがmmdetectionの本体には入っていないようなので、TTFNetのpython setup.py developから続きをやってみました。

3.1TTFNetのインストール

python setup.py developから再開すると、

(open-mmlab2) F:\Users\sounansu\Anaconda3\ttfnet2>python setup.py develop
.....
sigmoid_focal_loss_cuda.obj : error LNK2001: 外部シンボル ""public: long * __cdecl at::Tensor::data<long>(void)const " (??$data@J@Tensor@at@@QEBAPEAJXZ)" は未解決です。
build\lib.win-amd64-3.7\mmdet\ops\sigmoid_focal_loss\sigmoid_focal_loss_cuda.cp37-win_amd64.pyd : fatal error LNK1120: 1 件の未解決の外部参照
error: command 'C:\\Program Files (x86)\\Microsoft Visual Studio\\2017\\Community\\VC\\Tools\\MSVC\\14.16.27023\\bin\\HostX86\\x64\\link.exe' failed with exit status 1120

ここまでは、TTFNetのgithubがmmdetectionのと同じならば、本当ならば、うまくいくはずですが、TTFNetへfolkしたあと、mmdetectionの修正を追随していないようなので、mmdetection側のissueであるCompile successfully on Windows. #1088の

edit mmdetection/ops/masked_conv/src/masked_conv2d_kernel.cu`
find and replace: long -> int64_t

edit mmdetection/ops/sigmoid_focal_loss/src/sigmoid_focal_loss_cuda.cu
find and replace: long -> int64_t
in line 112, add type casting to values
dim3 grid(std::min(THCCeilDiv((long)losses_size, 512L), 4096L))
to
dim3 grid(std::min(THCCeilDiv((int64_t)losses_size, (int64_t)512L), (int64_t)4096L));
in line 150, add type casting to values
dim3 grid(std::min(THCCeilDiv((long)d_logits_size, 512L), 4096L));
to
dim3 grid(std::min(THCCeilDiv((int64_t)d_logits_size, (int64_t)512L), (int64_t)4096L));

をコードに反映されればいいのです。
mmdetection側は、この2.,3.はgitに反映されているのですが、TTFNetのgitにはされていないので、この2つのファイルをmmdetectionからとってきてTTFNetのフォルダに入れちゃいましょう。

(open-mmlab2) F:\Users\sounansu\Anaconda3>copy mmdetection\mmdet\ops\masked_conv\src\masked_conv2d_kernel.cu ttfnet\mmdet\ops\masked_conv\src\masked_conv2d_kernel.cu
ttfnet\mmdet\ops\masked_conv\src\masked_conv2d_kernel.cu を上書きしますか? (Yes/No/All): Y
        1 個のファイルをコピーしました。

(open-mmlab2) F:\Users\sounansu\Anaconda3>copy mmdetection\mmdet\ops\sigmoid_focal_loss\src\sigmoid_focal_loss_cuda.cu ttfnet\mmdet\ops\sigmoid_focal_loss\src\sigmoid_focal_loss_cuda.cu
ttfnet\mmdet\ops\sigmoid_focal_loss\src\sigmoid_focal_loss_cuda.cu を上書きしますか? (Yes/No/All): y
        1 個のファイルをコピーしました。

ではもう一度、python setup.py developを実行しましょう。

(open-mmlab2) F:\Users\sounansu\Anaconda3\ttfnet2>python setup.py develop
...
Using f:\users\sounansu\anaconda3\envs\open-mmlab2\lib\site-packages\decorator-4.4.0-py3.7.egg
Finished processing dependencies for mmdet==1.0rc0+99e9a6e

上手くいったっぽい。というかmmdetectionで実行したのと同じメッセージのようです。

3.2 TTFNetウエイトデータのダウンロード

TTFNetのREADME.mdのinferenceにあるLinkをクリックするとウエイトデータが得られるのでダウンロードしましょう。せっかくですので、全部取ってきましょう。とってきたデータはTTFnetのフォルダに入れておきましょう。また、mmdetectionに作ったcocoのデータが入っているdataフォルダと同じものをTTFnetのフォルダにも作っておきましょう。

4.TTFNetのインファレンス

　早速インファレンスを実行してみましょう。

(open-mmlab2) F:\Users\sounansu\Anaconda3\ttfnet2>set PYTHONPATH=F:\Users\sounansu\Anaconda3\ttfnet2

(open-mmlab2) F:\Users\sounansu\Anaconda3\ttfnet2>python tools\test.py configs\ttfnet\ttfnet_d53_1x.py ttfnet53_1x-4811e4.pth --show
loading annotations into memory...
Done (t=0.79s)
creating index...
index created!
The model and loaded state dict do not match exactly

unexpected key in source state_dict: bbox_head.shortcut_layers.0.shortcut.weight, bbox_head.shortcut_layers.0.shortcut.bias, bbox_head.shortcut_layers.1.shortcut.weight, bbox_head.shortcut_layers.1.shortcut.bias, bbox_head.shortcut_layers.2.shortcut.weight, bbox_head.shortcut_layers.2.shortcut.bias

[                                                  ] 0/5000, elapsed: 0s, ETA:

ウインドウがでて

のようなものが出れば正解です。set PYTHONPATHを忘れずに。

5.トレーニング

　ここまできたら、勝ったも同然。トレーニングをしましょう。dataフォルダが正しければちゃんと動きます。もしうまくいかなかったらdataフォルダを見直してください。

(open-mmlab2) F:\Users\sounansu\Anaconda3\ttfnet2>python tools\train.py configs\ttfnet\ttfnet_r18_1x.py
2019-10-06 19:37:25,824 - INFO - Distributed training: False
2019-10-06 19:37:26,388 - INFO - load model from: modelzoo://resnet18
2019-10-06 19:37:26,704 - WARNING - The model and loaded state dict do not match exactly

unexpected key in source state_dict: fc.weight, fc.bias

loading annotations into memory...
Done (t=23.91s)
creating index...
index created!
2019-10-06 19:37:59,831 - INFO - Start running, host: sounansu@sounansu-PC, work_dir: F:\Users\sounansu\Anaconda3\ttfnet2\work_dirs\ttfnet18_1x
2019-10-06 19:37:59,831 - INFO - workflow: [('train', 1)], max: 12 epochs
2019-10-06 19:40:25,909 - INFO - Epoch [1][50/7330]     lr: 0.00445, eta: 2 days, 23:16:16, time: 2.919, data_time: 1.341, memory: 5856, losses/ttfnet_loss_heatmap: 4.4741, losses/ttfnet_loss_wh: 4.8743, loss: 9.3484

うごいたことには動いたのですが、
eta: 2 days, 23:16:16, time:
2日以上？？？。タスクマネージャを見ると16GBしかメモリがないのに、15GB以上使っていそうで、いわゆるスワップが発生しているようです。
　いよいよ、CPU＋MB＋メモリの載せ替え時期が来たようです。。。。

終わりに

　TTFNetどうでしたでしょうか？たしかに、mAPの値とかはSOTAより劣っていますが、学習時間が短いというのは、比較的小さなGPUしかもっていない人には優しいといえなくもないです。PCのメインメモリが少ないため、私の環境では残念な結果になってしまいましたが、そもそも16GBは、前からしょぼすぎると思っていたので、これがいい機会かなと思います。（いまさらDDR3を32GBにはしたくないので、MB+CPU+MEM+SSDというフルセットでの変更になるでしょう。。。。
　では、またWindowsでInferenceやTrainingができるものがありましたら、紹介していきます。