{{unknown}}

1.chang the format of FLICK30

2.make the model with the FLICK30 dateset, the model is running now

1.recurrent inception+Transformer .

2.Research on encoder how to make innovation on inception or FCNN.

1.CBAM+CNN实现

2. Make a framework of FCNN+Tranformer

1.Read paper: 视觉问答中的多模态信息融合_庞章阳.pdf

注意力机制海洋场景图像理解算法_邬满.pdf

融合空间注意力机制的图像语义描述算法_郭列.pdf

基于多模态注意力机制的图像理解描述新方法_李学明.pdf

2.recurrent "pytorch-tutorial-image-caption "

3.make a demo on RNN、 LSTM

1.read paper：UNITER:UNIVERSAL IMAGE-TEXT REPRESENT LEARNING

https://mountain.elcs.kyutech.ac.jp/gao/read-paper/-/blob/master/UNITER_UNiversal_Image-TExt.pdf

2.read paper：

Show and Tell: Lessons learned from the 2015
MSCOCO Image Captioning Challenge

3.recurrent “2”

1. It took nearly a week to move, buy and tidy up
2. read paper： Ans wer_Questions_with_Right_Image_Regions_A_Visual.pdf

3.learn the code of VL-BERT

1.vl-bert-codeを研究

1.pytochを勉強する
2.論文を読む：Unifying_Vision-and-Language_Tasks_via_Text_Generation.pdf（https://mountain.elcs.kyutech.ac.jp/gao/read-paper/-/blob/master/Unifying_Vision-and-Language_Tasks_via_Text_Generation.pdf）
3.VL-bert 実行に成功しました

1.論文を読むpixel-bert（https://mountain.elcs.kyutech.ac.jp/gao/read-paper/-/blob/master/pixel-bert.pdf）

2.VL-bert 実行に成功しました，前トレーニングモデルを下流タスクにどのように適用するかを学ぶ。