HGA复现

项目来源：GitHub - Jumpin2/HGA: Reasoning with Heterogeneous Graph Alignment for Video Question AnsweringReasoning with Heterogeneous Graph Alignment for Video Question Answering - GitHub - Jumpin2/HGA: Reasoning with Heterogeneous Graph Alignment for Video Question Answeringhttps://github.com/Jumpin2/HGA

Pytorch code of Reasoning with Heterogeneous Graph Alignment for Video Question Answering.

Requirements

Python 3.6
Pytorch 1.1

Data

datasets：

「dataset」https://www.aliyundrive.com/s/mLg6FxZj1Q1 点击链接保存，或者复制本段内容，打开「阿里云盘」APP ，无需下载极速在线查看，视频原画倍速播放。

feats：HGA中的feat数据-深度学习文档类资源-CSDN下载

Vocabulary：

python3的，HGA中的feat数据-深度学习文档类资源-CSDN下载

Pre-trained Model

count：HGA中的feat数据-深度学习文档类资源-CSDN下载

 Test

main.py替换count模型名，运行

python main.py  --test --task Count --num_workers 2 --batch_size 64 

Train

python main.py --tount --num_workers 2 --batch_size 64 --lr 0.0001 --model 7 --dropout 0.3 --change_lr none --ablation none  --save