Introduction

Multiple idcard detection

Requirements

pip install requirements.txt

Demo

cd src/

CUDA_VISIBLE_DEVICES=0 python demo_classes.py
                            --model_path {MODEL_PATH}
                            --image_dir {IMAGE_DIR}
                            --output_dir {OUTPUT_DIR}
                            [--save_heatmap]
                            [--save_paf]

eg: CUDA_VISIBLE_DEVICES=0 python demo_classes.py --model_path 1000.ckpt --image_dir abc/ --output_dir def/

Quickstart

1. Dataset

Structure data root should be:

{DATA_ROOT}
|-- data
    |-- train
        |-- images
        label.csv
    |-- val
        |-- images
        |-- label.csv

label.csv

filename,width,height,class,tlx,tly,trx,try,brx,bry,blx,bly
a.jpg,624,1001,0,572,91,577,871,75,934,46,45
b.jpg,600,800,123,1,123,456,123,456,456,123,456
...

2. Training

1: Modify config.py

BATCH_SIZE: batch_size when train and evaluation

PRETRAINED: use pretrained ckpt or not

PRETRAINED_PATH: path to pretrained ckpt

MODEL_DIR: folder save ckpt while training

NUM_CLASSES: number of object classes (note: each classes must have same joint)

LABEL_MAP: label map of object, must same as class in file label.csv

2: Training

CUDA_VISIBLE_DEVICES=0 python main.py