A neural network to detect the total price printed on a super-market receipt. The goal is to create an real-time and accurate model that can be integrated into mobile applications.
I currently maintaining a dataset for training and validation of the model that contains 35 images. I collected these supermarket receipts, took pictures of them and labeled them respectively. I will add more and more labeled receipts in the future since 35 samples is too few for a robust model. When the dataset is big enough I may provide a link to download it. In the meantime I crop and scale the samples randomly to augment more data.
When executing main.py, you can use the following command-line options
| option | type | description | default |
|---|---|---|---|
--list-devices |
flag | Lists available processing units (GPU's, CPU's, etc.). Exits immediately after list is printed. | False |
--model_name |
string | The model's name needed for restoring and saving checkpoints. | "default" |
--config |
string | Filepath to the JSON configuration file. If the file doesn't exists under the specified path, a blank configuration file will be generated automatically | "config.json" |
--continue |
flag | If present it will try to restore the latest checkpoint (model_name determines which model will be restored) |
False |