huyhoang17 / kuzushiji_recognition

[Late Submission] Solution for Kuzushiji recognition (Kaggle competition)

Kuzushiji Recognition

[Late Submission] Solution for kuzushiji recognition (kaggle competition)

• Link blog post: Building OCR module for Kuzushiji recognition

Segmentation model

• Unet with custom resnet-based backbone

Evaluate on detection model

Classification model

• Baseline model for kuzushiji character recognition

• Number of classes: 3422

Command

• Clone repository

git clone https://github.com/huyhoang17/kuzushiji_recognition
cd kuzushiji_recognition

• Install some prerequisite libs

pip install -r requirements.txt

• [Optional] Install Git LFS and pull model files, follow by this tutorial

• Open kuzu_tfserving.config on editor, change base_path of 2 models to absolute path to each sub-folder

# change this line
base_path: '/home/phan.huy.hoang/workspace/projects/kaggle_kuzushiji/model_server/kuzu_segment'

# to
base_path: '/absolute-path-to-root-folder/model_server/kuzu_segment'

• Install tensorflow_model_server from this link

• Run tensorflow model server

tensorflow_model_server --port=8500 --rest_api_port=8501 --model_config_file=/absolute-path-to-kuzu-tfserving.config

• Test detection & recognition model

python3 src/grpc_infer.py

• Check result image in assets folder

Result

Result 2

• I tested on Chữ_Nôm and used pre-trained detection model from kuzushiji dataset

Đại Việt sử ký toàn thư

Chùa Thắng Nghiêm

Lục Vân Tiên

TODO

• Add pytorch code

• Update docker / docker-compose

Other issue

• Move Git LFS tracked file under regular git: git-lfs/git-lfs#3026 (comment)

Contact

• If you find this repo useful, please star the project to let people know that it's reliable ⭐⭐⭐ Thank you!

• For more information, please contact me at email address: [email protected]

kuzushiji_recognition's Issues

debug_error_string = "{"created":"@1610383278.487313200","description":"Failed to pick subchannel","file":"src/core/ext/filters/client_channel/client_channel.cc","file_line":4133,"referenced_errors":[{"created":"@1610383278.487310096","description":"failed to connect to all addresses","file":"src/core/ext/filters/client_channel/lb_policy/pick_first/pick_first.cc","file_line":397,"grpc_status":14}]}" >

can I have you ?