1.INSTALLATION

• Ubuntu 18.04.5 LTS
• CUDA 11.2
• Python 3.7.5
• Training PC: 1x RTX3090 (or any GPU with at least 24Gb VRAM), 32GB RAM.
• python packages are detailed separately in requirements.txt

$ conda create -n envs python=3.7.5
$ conda activate envs
$ pip install -r requirements.txt
$ pip install torch==1.10.1+cu113 torchvision==0.11.2+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html

2.DATA

• Kalapa dataset.
• Vietnamese address dataset https://github.com/thien0291/vietnam_dataset
• Vietnamese poems corpus: https://huggingface.co/datasets/phamson02/vietnamese-poetry-corpus
• Synthetic data. Download my generated data here (https://drive.google.com/file/d/1FfjVZqNRZGExZjZmqzWk_L3QDQS-20Bl/view?usp=sharing)
• Or following the commands below to re-generating it. (make sure poems_dataset.csv and vietnam_dataset are inside synthetic_data/)

$ cd synthetic_data  
$ python gendata_address.py    
$ python gendata_aug.py  
$ python gendata_poems.py  
$ cd ..  
$ python prepare_ext_data.py  

• Folder structure before executing training
  ├── training_data
  │ ├── images
  │ ├── annotations
  ├── synthetic_data
  │ ├── address
  │ ├── aug
  │ ├── poems
  │ ├── ...
  ├── configs
  │ ├── b2_256_ptr_f5.py
  │ ├── b1_384_ptr_f5.py
  │ ├── ...
  ├── train.py
  ├── prepare_ext_data.py
  ├── train_ext3.csv
  ├── train_folds.csv
  ├── ...

3.TRAINING

• Pretrained models on synthetic data.

$ python train.py -C b2_256_ptr_f5  
$ python train.py -C b1_384_ptr_f5  

• Fine-tune models on real data.

$ python train.py -C b1_384_f5  
$ python train.py -C b2_256_f5  

4.INFERENCE

• Refer to submitted notebook
• Weights: https://drive.google.com/file/d/1CsWm4RT6tW-u8A0avVPiooZv9G_Lxpm6/view?usp=sharing