HI19

HI-19 is a system developed for identifying and distinguishing peoples' handrwritings. This is the first such product.

Generally

How does HI-19 work?

1. First, the received photo is segmented into words
2. Afterwards, we need to fit the model. HIModel из model
3. To work with HI-19 API we recommend to use RunEnv for run-time and FitEnv for fitting, both from hypo2.api

Structure

• config.py - project configuration file
• model.py - implementation and functions of the HIModel
• api.py

Model

First of all

First, we configure the project. Create file myconfig.py and insert the following code:

from hypo2.config import RunConfig

class MyConfig(RunConfig):
    BACKUP_DIRECTORY = ???
    MODEL_PATH = ???

Substitute ??? with either the appropriate path or None. BACKUP_DIRECTORY is a path to the directory, where backup models will be saved. MODEL_PATH - a path to the final model.

Then, let us create the config variable. It is simple:

from myconfig import MyConfig
config = MyConfig()

So, you can change some settings right in the run-time. To have a look at the configs just type

print(config)

Fitting

The dataset we have should be structured in the following way:

[
  [
  "0/0.jpg",
  "0/1.jpg",
  "0/2.jpg",
  ],
  [
  "1/0.jpg",
  "1/1.jpg"
  ],
  [
  "2/0.jpg"
  ]
]

In other words, every item of the array is an array of links to photos of one class. You have to change the config variable, at least, CLASS_COUNT. Let us say we already have dataset. Imporint api

from hypo2.api import FitEnv

FitTime enables to fit the model easily. Creating an instance of FitEnv:

fitenv = FitEnv(config)

The environment is ready to fit. FitEnv will create a model, process the input, and set the system. Let us fit the model:

model = fitenv.fit(dataset, verbose=True, plot=False)

verbose is whether to output epoch results. plot is whether to plot loss diagram. model is the result of fitting.

Model methods

You can save your model via

model.save()

The path the model will be saved to is config.MODEL_PATH

If you want to do it manually, specify the path (deprecated method)

model.saveto("D:/HI.19")

To open the model, specify it in config.MODEL_PATH and then create an instance.

config.MODEL_PATH = "D:/HI.19"
model = HIModel(config)

If you want to reopen the model

model.open()

Or use a deprecated method:

model.openfrom("HI.19")

Usage

Importing and creating RunEnv

from hypo2.api import RunEnv
runenv = RunEnv(config)

Let us try to get a Center by photo:

image = runenv.open_image("johns_text.jpg")
cw = runenv.get_center(image)
assert cw is not None, "0 words found"
center, weight = cw

Where center is a vector of length config.FEATURES_COUNT, and weight is an integer, that is equal to the total amound of found words. To compare two images and get the difference between them, use differ

distance = runenv.differ(image1, image2)

If you already have centers, then you can find out the distance between them using function

distance = runenv.dist(john_center, mary_center)