[TOC]
. vector_ Cache (hidden file): Glove vector storage address
captioning: The address where the text generated by the video description is stored
checkpoints: The address of the saved model file
data: Source video data
loader: Load data data
logs: Logs saved during model training, which can be used in conjunction with Tensorboardx
models: Model files
pycocoevalcap: Code for calculating indicators such as BLEU, CIDER, etc.
result: The address where the results of the trained model are stored
splits: The method of data cutting
config.py: Configuration file
run. py: Test model
train.py: training model+testing model
utils. py: Some public methods
- First, create a conda virtual environment through cyd.yaml (which stores the project's environment configuration)
Conda env create - f cyd.yaml
- After configuring the environment: directly run the train.py file (if the GPU graphics memory is insufficient, you can reduce the configuration through config)
- Generally no problem, ask me if you have any questions
Download link: https://pan.quark.cn/s/44049885ed0b
.vector_cache(隐藏文件):glove 向量存放地址
captioning:视频描述生成的文本存放的地址
checkpoints:保存的模型文件的地址
data:源视频数据
loader:加载data数据
logs:训练模型时保存的日志,可配合 tensorboardx 一起使用
models:模型文件
pycocoevalcap:BLEU、CIDER…等指标计算的代码
result:训练好的模型的结果存放的地址
splits:数据切割的方式
config.py:配置文件
run.py:测试模型
train.py:训练模型+测试模型
utils.py:一些公共方法
- 首先通过 cyd.yaml(里面存放了项目的环境配置) 创建一个 conda 虚拟环境
conda env create -f cyd.yaml
- 配置好环境后:直接运行 train.py 文件(如果GPU 显存不够,可以通过 config 减小配置)
- 一般没问题,有问题问我