bee992 / Text-Manipulation-Classification Public

Notifications You must be signed in to change notification settings
Fork 2
Star 5

8th solution for "Tianchi-ICDAR 2023 DTT in Images 1: Text Manipulation Classification" competition.

5 stars 2 forks Branches Tags Activity

Notifications

Name		Name	Last commit message	Last commit date
Latest commit History 2 Commits
__pycache__		__pycache__
README.md		README.md
augmentation.py		augmentation.py
dataset.py		dataset.py
lookahead.py		lookahead.py
model.py		model.py
nextvit.py		nextvit.py
run.py		run.py
run.sh		run.sh
run_aug.py		run_aug.py
run_aug_0307.py		run_aug_0307.py
run_b6_0308.py		run_b6_0308.py
run_nextvit_0308.py		run_nextvit_0308.py
submit.py		submit.py
test.csv		test.csv
train.sh		train.sh
utils.py		utils.py

Repository files navigation

Text-Manipulation-Classification

8th solution for "Tianchi-ICDAR 2023 DTT in Images 1: Text Manipulation Classification" competition.

技术路线

利用EfficientNet B6作为特征提取器，将提取到的特征用MLP做二分类
同理，利用NextViT作为特征提取器
融合两个模型的结果，最终复赛排名第8

提分点

数据预处理很重要：由于输入图像size不一致，输入时需要resize，此时resize的大小会对模型的指标带来很大影响
数据增强很重要，在其他CV分类任务中常用的一些增强手段在该任务中应慎用or使用较小的阈值来约束，保证训练数据的稳定性
外部数据集预训练

About

8th solution for "Tianchi-ICDAR 2023 DTT in Images 1: Text Manipulation Classification" competition.

Report repository

Releases

No releases published

Packages

No packages published

Languages