工作日常:打开Github开始新的一天
- SAM
- RobustScanner: Dynamically Enhancing Positional Clues for Robust Text Recognition
- Translating Math Formula Images to LaTeX Sequences Using Deep Neural Networks with Sequence-level Training
- ConvMath: A Convolutional Sequence Network for Mathematical Expression Recognition
- wzlxjtu/PositionalEncoding2D
- im2markup from Image-to-Markup Generation with Coarse-to-Fine Attention
- LaTeX-OCR pix2tex: Using a ViT to convert images of equations into LaTeX code.
- image-to-latex transformer
- im2latex: Seq2Seq model with Attention - Beam Search for Image to LaTeX, similar to Show, Attend and Tell and Harvard's paper and dataset.
- im2latex: Pytorch implemention of Deep CNN Encoder - LSTM Decoder with Attention for Image to Latex
- math-formula-recognition WAP
- Pytorch-Handwritten-Mathematical-Expression-Recognition WAP
- BTTR: BiTransformer
- CAN
- CoMER
- Master-Ocr
- Master-Table
- SAN
- 文章
- 论文
- Multi-Scale Attention with Dense Encoder for Handwritten Mathematical Expression Recognition
- Watch, attend and parse: An end-to-end neural network based approach to handwritten mathematical expression recognition
- MASTER: Multi-Aspect Non-local Network forScene Text Recognition
- Handwritten Mathematical Expression Recognition with Bidirectionally Trained Transformer
- Recognizing handwritten mathematical expressions via paired dual loss attentionnetwork and printed mathematical expressions
- DenseWAP-TD
- Handwritten Mathematical Expression Recognition with Bidirectionally Trained Transformer
- gordicaleksa/pytorch-GAT 图注意力机制
- External-Attention-pytorch
- Image-Local-Attention
- 文章
- 一文看懂 Attention(本质原理+3大优点+5大类型)
- 自注意力机制的PyTorch实现
- PyTorch教程: seq2seq机器翻译及代码实现
- 序列到序列 (Seq2Seq) 和注意力机制 (Attention Machanism)
- 用pytorch实现RNN(两种构造RNN的方法;序列到序列的训练),PyTorch,方式
- 动画图解Attention机制,让你一看就明白
- Visualizing A Neural Machine Translation Model (Mechanics of Seq2seq Models With Attention)
- Attention Craving RNNS: Building Up To Transformer Networks
-
数据合成
-
OCR
-
数据增广
-
常用数据集网站
- CHROME
- Pix2seq: A Language Modeling Framework for Object Detection 语言模型搭建检测
- DefTruth/lite.ai C+- 检测部署
- PeizeSun/OneNet
- Yolov5_DeepSort_Pytorch
- 文本检测
- 表格检测
- ATSS 目标检测采样策略
- WeightedBoxesFusion 目标检测集成方法
- SoftTeacher
- 单阶段
- 部署
- 多方向
- 目标跟踪
- 基于概率分布图的任意形状文本实例分割和检测方法(有源码
- [CVPR2022] 端到端的场景文字检测与版面分析统一框架
- [CVPR 2022] 特征采样与分组:基于Transformer的场景文字检测方法
- 滑铁卢大学:使用transformer的任意形状文本检测
- ATSS
- 目标检测评价指标详解
- Weighted boxes fusion
- 涨点神器!南航提出AFF:注意力特征融合,即插即用!可用于分类、检测和分割等
- 在目标检测被“遗忘”领域进行探索后,百度开源最新力作UMOP:即插即用、无痛涨点
- 竞赛经验|Kaggle竞赛中使用YoloV5将物体检测的性能翻倍的心路历程
- 涨点神器!SIoU:目标检测的新损失函数,提高准确性和训练速度!
- 目标检测正负样本区分策略和平衡策略总结
- ashi|小目标Trick | Detectron2、MMDetection、YOLOv5都通用的小目标检测解决方案
- EDA: Exploratory Data Analysis
- TTA
- softmax-nms
- OHEM
- SWA
- RoIAlign
- 多尺度训练、推理
- 根据batchsize的大小适当调整bn的类型
- 目标检测比赛提高mAP的方法
- 弯曲文本检测策略
- [hiertext](https://github.com/google-research-datasets/hiertext)
- icdar
- ctw
- total text
- openimage
- cocotext
- 数据增广
- 梯度正则
- 学习率调整方法
- 权重正则
- 多卡训练
- 注意力机制
- 激活函数
- Center Loss
- SWA
- 自集成(swa)/自蒸馏(self-distillation)
- Test Time Augmentation(TTA)
- Ensemble-Pytorch
- MEAL-V2 KL散度以及对抗损失
- 数据平衡
-
移动到TODO.md
- Pytorch的一些基本操作
- ritchieng/the-incredible-pytorch 教程
- 长尾问题
- facebookresearch/FixRes trick
- imankgoyal/NonDeepNetworks ParNet
- assafshocher/ResizeRight
- kornia/kornia
- Learning to Resize Images for Computer Vision Tasks
- pytorch/libtorch
- pytorch/examples
- pytorch-optimizer
- victoresque/pytorch-template
- pytorch-metric-learning
- 自定义操作torch.autograd.Function
- pytorch-loss
- BatchNorm 很重要,也很有效
- 模型加速
- 无监督
- Pytorch 入门
- KL散度理解以及使用pytorch计算KL散度
- AutoClip: Adaptive Gradient Clipping
- timm
- Test Time Augmentation(TTA)
- Ensemble-Pytorch
- MEAL-V2 KL散度以及对抗损失
- 数据平衡
- CNN
- lightly 弱监督
- mit-han-lab/once-for-all
- IntelLabs/distiller
-
vscode ssh配置文件设置代理
-
Ray
-
wandb
-
Time
-
MarkDown
-
gpustat
-
lsof -i:port 查看端口使用情况
-
Git
- commit之后,想撤销commit
- git checkout
- git reset [--mixed [--soft [--hard ]]]
- git stash
- git stage
- commit之后,想撤销commit
-
命令行工具
-
深度学习/环境配置
-
穿透
- frp
- cploar
- ssh
- zerotier
- clash
-
Win
- win 自启动目录 shell:startup
-
自建云盘
-
Github
-
订阅
-
chrome
-
brave
-
代理
- switch-hosts
- SwitchyOmega
-
GPU主机市场
-
vscode
-
可视化
-
Three mysteries in deep learning: Ensemble, knowledge distillation, and self-distillation
-
Sequence-to-Sequence Contrastive Learning for Text Recognition
-
自蒸馏
- Handwriting-Transformers
- Transformer_STR
- handwriting-synthesis
- Generating-Sequences-With-Recurrent-Neural-Networks
- CyberAgentAILab/derendering-text
- gidariss/FeatureLearningRotNet RotNet
- RecycleNet: An Overlapped Text Instance Recovery Approach 文本图像分离
- yizhiwang96/deepvecfont 手写生成
- Full Page Handwriting Recognition via Image to Sequence Extraction --transfoemer
- NRTR, a No-Recurrence Seq2Seq Model for Scene Text Recognition Belval/NRTR
- From Two to One: A New Scene Text Recognizer with Visual Language Modeling Networ VisionLAN
- SpellGCN: Incorporating Phonological and Visual Similarities into Language Models for Chinese Spelling Check 基于BERT的中文纠错
- microsoft/trOCR transformer 关键信息抽取
- TextStyleBrush
- Jianf-Wang/GRCNN-for-OCR
- clovaai/synthtiger
- 评估
- Enhanced CTC Loss
- icdar 评估
- sightseq
- mmocr
- DAVAR-Lab-OCR
- ACE LOSS
- rflearning
- deep-text-recognition-benchmark
- deep-text-recognition-benchmark --vit ocr
- vedastr
- KISS: Keeping it Simple for Scene Text Recognition.
- mtl-text-recognition
- textocr facebook
- 弱监督
- 文档矫正
- TODO
- 实战
学习相关的部署操作以及框架应用
- 文章
- PDF-Resume-Information-Extraction
- Vision Transformer
- x-transformers
- longformer
- huggingface transformers
- Gsunshine/Enjoy-Hamburger
- fast-transformers
- NVIDIA FasterTransformer
- multidim-positional-encoding
- lucidrains/HTM-pytorch Hierarchical Transformer Memory
-
help
-
格式处理相关
- pprint
- rich
- python-tabulate
- prettytable
- Typer
- click
- args
- icecream
- mitmproxy
- requests
- subprocess
- seleinum
-
常用内置函数
- int
- bin
- oct
- hex
- ord
- chr
-
字符串处理相关
- unicodedata
- string
- string.translate
- re
- Flashtext
- 中文字符串正则-\u4e00-\u9fa5
- Levenshtein
-
占用内存的分析
- torch
- tensor.storage()
- numpy
- ndarray.nbytes
- python 对象
- sys.getsizeof()
- psutils
- torch
-
python 持久化
- pickle
- torch.save
-
常用的数据存储数据库
- lmdb
-
采样
- random
- choice
- sample
- itertools
- product(p, q, … [repeat=1])
- permutations(p[, r])
- combinations(p, r)
- combinations_with_replacement(p, r)
- random
-
Debug
-
Remote Debug:****
{
"name": "Python: Dist Current File",
"type": "python",
"justMyCode": false,
"request": "attach",
"connect": {
"host": "localhost",
"port": 5678
}
// python -m debugpy --listen 5678 --wait-for-client
}
-
编译/运行库相关
- ldd
- cmake
- ccmake
- make
- bazel 编译工具
-
OpenCV
-
Ubuntu
- zsh
- fzf
- pip
- usermod / passwd 修改默认目录
- chmod / chown 修改文件权限
- gpustat 查看显卡的利用率
- 查看系统版本
cat /etc/issue
- 查看cuda安装成功
nvcc -V
- CUDA环境配置文件
-
TensorRT
- MayankSingal/PyTorch-Image-Dehazing
- xinntao/Real-ESRGAN-ncnn-vulkan
- edge-SR: Super-Resolution For The Masses 一层卷积进行超分
- thunil/TecoGAN
- Cross-Camera Convolutional Color Constancy
- NVlabs imaginaire
- MIMO-UNet
- Real-ESRGAN
- BasicSR-examples
- xindongzhang/ECBSR
- xiaomi-automl/FALSR
- Thmen/EGVSR
- Restomer
- AirNet
- 英语
- 正则
- ziishaned/learn-regex
- cdoco/common-regex
- vi3k6i5/flashtext
- re
- 中文正则 [\u4e00-\u9fa5]
- C++
- 论文-Search
- ML-YouTube-Courses
- 莫凡
- zh-v2.d2l.ai
- awesome-productivity-cn
- learnopencv
- learning-machine: A handbook for ML built on answers.
- full-stack-deep-learning/fsdl-text-recognizer-2021-labs
- UI开发
- 小说
- 万域之王
- 斗罗大陆
- 明朝那些事
- 星门
- 一念永恒
- 少年歌行
- 漫画
- 妖怪名单
- 传武
- 大神仙
- 公众号
- 量子位
- 52CV
- 常用网站
- Github
- 知乎
- enmmmm
- ssss
- CVPR2022 UG2+ 赛道三 冠军
- 科大讯飞 2022多学科公式识别 冠军
- 科大讯飞 2023复杂场景公式识别 冠军
- TIE2022 端到端 第六
- sss
- sss