Skip to content

Latest commit

 

History

History
73 lines (53 loc) · 5.58 KB

README.md

File metadata and controls

73 lines (53 loc) · 5.58 KB

Counting Real Money by Fake Money Data

1. Preprocess coin images

Original coin images

python make_seed.py 
      --input_path=data/coin_images 
      --output_path=data/coin_seeds 
      --thresh=120 --morph_iteration=3

After running the above command, seed coin images are obtained:

2. Create fake image data for training object detection model

Background images + Coin images -> Image with coins

Some background images

Now we have backgrounds and seed coins, we will synthesize fake data by randomly sawing seed coins on backgrounds. Run the command:

python synthesize.py --background_path=data/background
                     --seed_path=data/coin_seeds
                     --num_images=400

We will obtains 400 fake images like this:

and the corresponding bounding box annotation:

t_207.jpg 847,530,1024,706,7 576,751,755,928,6 406,197,612,402,5 110,233,273,403,6 666,178,869,380,4 325,512,498,684,7 1040,155,1258,379,3 132,659,299,828,7 1120,433,1289,606,6
t_213.jpg 990,496,1104,609,5 565,585,662,681,7 684,391,811,518,2 121,514,258,649,1 1185,826,1303,945,4 1106,303,1231,428,2 502,700,597,795,6 1168,613,1298,744,2 211,96,353,237,1 541,196,667,322,2 1107,115,1241,250,1 369,662,486,780,5
t_356.jpg 657,175,849,368,1 241,431,412,600,2 425,678,615,872,0 990,306,1185,502,1 262,104,396,239,6 1057,670,1199,810,6 818,555,1014,751,1 173,793,314,935,6 442,367,577,502,7

3. Convert image data into TFrecord format for training with Tensorflow Object Detection API

python make_tfrecord.py

4. Training

python train_model.py --model_dir=models/intraining_models/ssd 
                      --pipeline_config_path=config/pipeline_ssd.config

5. After training, count coins on real images:

python count_coint.py --image_path=data/real_coin_images/c.jpg 
                      --model_path=models/exported_models/ssd

The output

6. To download

the fake data: https://drive.google.com/drive/folders/1oC-utM77Pgzykd1lMt8RlUmNsmyoMThq?usp=sharing

the trained ssd model: https://drive.google.com/drive/folders/10niXINYoADLxuuo6W-PI_aXzZNh02h6J?usp=sharing