SumasenTrainer/README.md

# Sumasen Trainer
---
### Train and create a new Model

1) Select New from left menu
![initial](documents/initial.png) 

2) Enter image folder where images are saved in sub folder as their category

**This will assign a outpu folder to arrage images, you can modify if required**
![output folder](documents/output_folder.png)

Eg:
`image_classifier/dataset/raw_data`
![select image folder](documents/select_image_folder.png)


3) Select input shape
This will be used to resize image into
![image shape](documents/image_shape.png)

4) Select Batch Size
   This is the number of image used to train as a batch
![batch size](documents/select_batch_size.png)

5) Click arrange images button
This will rearrange the images seected above and arrange them as train, test and validation sets
![arranges images](documents/arraged_images.png)

6) a) Enter a name for model, the model will be saved under previouly selected output folder
  b) also select number of epochs the training should run
![start training](documents/begin_training.png))

7) training progrss
![training progress](documents/training_progress.png)

8) Training completed.
   The model will be saved in the output folder
   ![training complted](documents/finished_training.png)


### Predicting a new image

1) Select "Existing model" from left pane

![existing model](documents/previous_model.png)

2) Enter a folder to list all models available (if a training session is done previously the output folder will be there automatically)
3) Click load model and select a model available
![select model](documents/select_model.png)

4) On the right pan select images shape that used for this model and a new image to predict,
![predict](documents/predicttion.png)

### Automated categorizer

#### How it works
- When an image is placed in a defined monitoring folder the categoriser will use defined trained model to categorise it and will move the image to a category sub folder of the defined target folder

- defining model, monitoring folder and target folder:
  - "params.dat" file have those vales