在复现ICAL实验过程中遇到一些问题

Kerston12138 commented 3 months ago

作者你好！我在复现你们项目的时候遇到一些基础的问题盼望您有空的解答

1、训练好的权重文件保存在哪里？ 2、怎么用训练好的权重去对单张图片做预测？ 3、对单张图片做预测的话怎么分别查看 ICAL 和 ICCM的预测结果？ 4、在model->decoder.py中 SCCM 这里主要的目的是什么，在论文中没有写。

qingzhenduyu commented 3 months ago

Hello, q1,q2: It is recommended to refer to the official Lightning documentation for learning: Lightning Documentation. For further guidance, please consult the official Lightning documentation. q3: You can simply refine the eval/test.py script for a few lines. q4:" SCCM is a Transformer Encoder designed to predict implicit tokens, and it has been renamed as ICCM in our paper.

Thanks

Kerston12138 commented 1 week ago

下午好师兄

我叫牟世奇，目前是23级硕士研究生 ,也在研究HMER任务，在研究中遇到几个疑惑，没能弄明白。所以给您写信，请师兄指教一下。

问题描述如下：

1· crohme数据集的vocab_size=114 但是我在data->crohme->dictionary.txt中只看到了110个字符,请问还有四个是什么字符呢？

2· 这些字符与目标序列的编码映射关系是什么？我尝试打印了tgt的内容发现如果按照dictionary的顺序映射并不能组成一个合理的公式。

期待您的回信感谢

------------------ 原始邮件 ------------------ 发件人: "qingzhenduyu/ICAL" @.>; 发送时间: 2024年5月27日(星期一) 晚上11:09 @.>; @.**@.>; 主题: Re: [qingzhenduyu/ICAL] 在复现ICAL实验过程中遇到一些问题 (Issue #3)

Hello, q1,q2: It is recommended to refer to the official Lightning documentation for learning: Lightning Documentation. For further guidance, please consult the official Lightning documentation. q3: You can simply refine the eval/test.py script for a few lines. q4:" SCCM is a Transformer Encoder designed to predict implicit tokens, and it has been renamed as ICCM in our paper.

Thanks

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

qingzhenduyu commented 1 week ago

你可以参照ICAL/ical/datamodule/vocab.py来找到另外四个字符；
如果你加上这4个字符，是不是就合理了？另外，建议直接indices2words函数来进行映射，会方便一点。祝好！ Jianhua Zhu

qingzhenduyu / ICAL

在复现ICAL实验过程中遇到一些问题 #3