meshed-memory transformer代码实现（绝对详细）

meshed-memory transformer代码实现

参考的官方代码：

GitHub - aimagelab/meshed-memory-transformer: Meshed-Memory Transformer for Image Captioning. CVPR 2020

---

克隆存储库并m2release使用文件创建 conda 环境environment.yml：

conda env create -f environment.yml
conda activate m2release

运行程序报错：

Collecting package metadata (repodata.json): done
Solving environment: failed

ResolvePackageNotFound: 
  - intel-openmp==2019.5=281
  - openssl==1.1.1=h7b6447c_0

参考链接中贴出的环境

将源代码中

- openssl==1.1.1=h7b6447c_0

加g，改为

- openssl=1.1.1g=h7b6447c_0

然后参考博客将intel-openmp=2019.5=281注掉：

#- intel-openmp=2019.5=281

重新create，产生各种冲突

The following specifications were found to be incompatible with your system:

  - feature:/linux-64::__glibc==2.27=0
  - feature:|@/linux-64::__glibc==2.27=0
  - cffi==1.13.2=py36h2e261b9_0 -> libgcc-ng[version='>=7.3.0'] -> __glibc[version='>=2.17']
  - cryptography==2.8=py36h1ba5d50_0 -> libgcc-ng[version='>=7.3.0'] -> __glibc[version='>=2.17']
  - cython==0.29.14=py36he6710b0_0 -> libgcc-ng[version='>=7.3.0'] -> __glibc[version='>=2.17']
  - libedit==3.1.20181209=hc058e9b_0 -> libgcc-ng[version='>=7.3.0'] -> __glibc[version='>=2.17']
  - libffi==3.2.1=hd88cf55_4 -> libgcc-ng[version='>=7.2.0'] -> __glibc[version='>=2.17']
  - mkl-service==2.3.0=py36he904b0f_0 -> libgcc-ng[version='>=7.3.0'] -> __glibc[version='>=2.17']
  - mkl_fft==1.0.15=py36ha843d7b_0 -> libgcc-ng[version='>=7.3.0'] -> __glibc[version='>=2.17']
  - mkl_random==1.1.0=py36hd6b4f25_0 -> libgcc-ng[version='>=7.3.0'] -> __glibc[version='>=2.17']
  - msgpack-python==0.5.6=py36h6bb024c_1 -> libgcc-ng[version='>=7.2.0'] -> __glibc[version='>=2.17']
  - ncurses==6.1=he6710b0_1 -> libgcc-ng[version='>=7.3.0'] -> __glibc[version='>=2.17']
  - openssl==1.1.1g=h7b6447c_0 -> libgcc-ng[version='>=7.3.0'] -> __glibc[version='>=2.17']
  - python==3.6.9=h265db76_0 -> libgcc-ng[version='>=7.3.0'] -> __glibc[version='>=2.17']
  - readline==7.0=h7b6447c_5 -> libgcc-ng[version='>=7.3.0'] -> __glibc[version='>=2.17']
  - spacy==2.0.11=py36h04863e7_2 -> libgcc-ng[version='>=7.2.0'] -> __glibc[version='>=2.17']
  - sqlite==3.30.1=h7b6447c_0 -> libgcc-ng[version='>=7.3.0'] -> __glibc[version='>=2.17']
  - thinc==6.11.2=py36hedc7406_1 -> libgcc-ng[version='>=7.2.0'] -> __glibc[version='>=2.17']
  - tk==8.6.8=hbc83047_0 -> libgcc-ng[version='>=7.3.0'] -> __glibc[version='>=2.17']
  - xz==5.2.4=h14c3975_4 -> libgcc-ng[version='>=7.2.0'] -> __glibc[version='>=2.17']
  - zlib==1.2.11=h7b6447c_3 -> libgcc-ng[version='>=7.3.0'] -> __glibc[version='>=2.17']

Your installed version is: 2.27

经百度，冲突原因是别人电脑和你的电脑的细节不一样，所以把所有第二个等号的内容删掉

dependencies:
  - _libgcc_mutex=0.1=main
  - asn1crypto=1.2.0=py36_0
  - blas=1.0=mkl
  - ca-certificates=2019.10.16=0
  - certifi=2019.9.11=py36_0
  - cffi=1.13.2=py36h2e261b9_0
  - chardet=3.0.4=py36_1003
  - cryptography=2.8=py36h1ba5d50_0
  - cython=0.29.14=py36he6710b0_0
  - dill=0.2.9=py36_0
  - idna=2.8=py36_0
  - intel-openmp=2019.5=281
  - libedit=3.1.20181209=hc058e9b_0
  - libffi=3.2.1=hd88cf55_4
  - libgcc-ng=9.1.0=hdf63c60_0
  - libgfortran-ng=7.3.0=hdf63c60_0
  - libstdcxx-ng=9.1.0=hdf63c60_0
  - mkl=2019.5=281
  - mkl-service=2.3.0=py36he904b0f_0
  - mkl_fft=1.0.15=py36ha843d7b_0
  - mkl_random=1.1.0=py36hd6b4f25_0
  - msgpack-numpy=0.4.4.3=py_0
  - msgpack-python=0.5.6=py36h6bb024c_1
  - ncurses=6.1=he6710b0_1
  - openjdk=8.0.152=h46b5887_1
  - openssl=1.1.1=h7b6447c_0
  - pip=19.3.1=py36_0
  - pycparser=2.19=py_0
  - pyopenssl=19.1.0=py36_0
  - pysocks=1.7.1=py36_0
  - python=3.6.9=h265db76_0
  - readline=7.0=h7b6447c_5
  - requests=2.22.0=py36_0
  - setuptools=41.6.0=py36_0
  - six=1.13.0=py36_0
  - spacy=2.0.11=py36h04863e7_2
  - sqlite=3.30.1=h7b6447c_0
  - termcolor=1.1.0=py36_1
  - thinc=6.11.2=py36hedc7406_1
  - tk=8.6.8=hbc83047_0
  - toolz=0.10.0=py_0
  - urllib3=1.24.2=py36_0
  - wheel=0.33.6=py36_0
  - xz=5.2.4=h14c3975_4
  - zlib=1.2.11=h7b6447c_3

改为

dependencies:
  - _libgcc_mutex=0.1
  - asn1crypto=1.2.0
  - blas=1.0
  - ca-certificates=2019.10.16
  - certifi=2019.9.11
  - cffi=1.13.2
  - chardet=3.0.4
  - cryptography=2.8
  - cython=0.29.14
  - dill=0.2.9
  - idna=2.8
  #- intel-openmp=2019.5=281
  - libedit=3.1.20181209
  - libffi=3.2.1
  - libgcc-ng=9.1.0
  - libgfortran-ng=7.3.0
  - libstdcxx-ng=9.1.0
  - mkl=2019.5
  - mkl-service=2.3.0
  - mkl_fft=1.0.15
  - mkl_random=1.1.0
  - msgpack-numpy=0.4.4.3
  - msgpack-python=0.5.6
  - ncurses=6.1
  - openjdk=8.0.152
  - openssl=1.1.1
  - pip=19.3.1
  - pycparser=2.19
  - pyopenssl=19.1.0
  - pysocks=1.7.1
  - python=3.6.9
  - readline=7.0
  - requests=2.22.
  - setuptools=41.6.0
  - six=1.13.0
  - spacy=2.0.11
  - sqlite=3.30.1
  - termcolor=1.1.0
  - thinc=6.11.2
  - tk=8.6.8
  - toolz=0.10.0
  - urllib3=1.24.2
  - wheel=0.33.6
  - xz=5.2.4
  - zlib=1.2.11

继续create，又报错：

pip._vendor.urllib3.exceptions.ReadTimeoutError: HTTPSConnectionPool(host='files.pythonhosted.org', port=443): Read timed out.

failed

CondaEnvException: Pip failed

分析原因是pip的镜像问题，参考博客将镜像设置默认

pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple

继续create，终于成功啦！！！！结果如下：

Successfully installed absl-py-0.8.1 cycler-0.10.0 cytoolz-0.9.0.1 dill-0.3.4 future-0.17.1 grpcio-1.25.0 h5py-2.8.0 kiwisolver-1.1.0 markdown-3.1.1 matplotlib-2.2.3 msgpack-0.6.2 multiprocess-0.70.9 numpy-1.16.4 pathlib-1.0.1 pathos-0.2.3 pillow-6.2.1 pox-0.2.7 ppft-1.6.6.1 protobuf-3.10.0 pycocotools-2.0.0 pyparsing-2.4.5 python-dateutil-2.8.1 pytz-2019.3 regex-2017.4.5 tensorboard-1.14.0 torch-1.1.0 torchvision-0.3.0 tqdm-4.32.2 ujson-1.35 werkzeug-0.16.0

done
#
# To activate this environment, use
#
#     $ conda activate m2release
#
# To deactivate an active environment, use
#
#     $ conda deactivate

激活环境，因为用的是miniconda，所以不要听上面建议的，而要用：

source activate m2release

成功激活！然后下载spacy文件，按照建议命令：

python -m spacy download en

安装成功！

Successfully built en-core-web-sm
Installing collected packages: en-core-web-sm
Successfully installed en-core-web-sm-2.0.0

    Linking successful
    /home/sui/.conda/envs/m2release/lib/python3.6/site-packages/en_core_web_sm
    -->
    /home/sui/.conda/envs/m2release/lib/python3.6/site-packages/spacy/data/en

    You can now load the model via spacy.load('en')

测试一下，成功！如下：

python 
Python 3.6.9 |Anaconda, Inc.| (default, Jul 30 2019, 19:07:31) 
[GCC 7.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import spacy
>>> spacy.load('en')

python train.py --exp_name m2_transformer --batch_size 50 --m 40 --head 8 --warmup 10000 --features_path /path/to/features --annotation_folder annotations