SGRF library development setup

Prerequisites

• Python 3.11
• PIP

Development

Installation

1. Create/activate virtual environment
2. Install required packages with pip install -r requirements.txt

Create new algorithm

To create a new algorithm, use algorithm creation script: ./scripts/generate_algorithm.py

Validate algorithms

To validate algorithms use scripts located in ./validation directory. Json files with validation results are located in ./validation/results directory

Usage

Import library

To use library in external project, use pip install sgrf.

Sample use cases

To predict gesture on selected image, run the code below. You can select desired algorithm by using values on ALGORITHM enum. Some algorithms require their own payload (e.g. hand coordinates or background image without hand). You can import specific payload from sgrf.algorithms.<alg>.<alg>_payload.

import cv2
from sgrf import classify
from sgrf.data.algorithm import ALGORITHM
from sgrf.models.image_payload import ImagePayload

image = cv2.imread("resources/image.jpg")
result = classify(algorithm=ALGORITHM.EID_SCHWENKER, payload=ImagePayload(image=image))

print(result)

To show image processed by the selected algorithm, run:

import cv2
from sgrf import process_image
from sgrf.data.algorithm import ALGORITHM
from sgrf.models.image_payload import ImagePayload

image = cv2.imread("resources/image.jpg")
processed_image = process_image(algorithm=ALGORITHM.EID_SCHWENKER, payload=ImagePayload(image=image))

cv2.imshow("Image", processed_image)
cv2.waitKey(0)
cv2.destroyAllWindows()

To learn your own algorithm's model (e.g. on other image base than ours), run:

import cv2
from sgrf import learn
from sgrf.data.algorithm import ALGORITHM
from sgrf.data.gesture import GESTURE
from sgrf.models.learning_data import LearningData

image = cv2.imread("resources/image.jpg")
acc, loss = learn(algorithm=ALGORITHM.EID_SCHWENKER, target_model_path="models",
                  learning_data=[LearningData(image_path="resources/image.jpg", label=GESTURE.FIVE)] * 10)

print(acc, loss)

Usage with Nextcloud

To use SGRF with Nextcloud BDGS, complete following steps:

1. Create ./env file.
2. Copy ./example.env contents to ./env file. Adjust settings with credentials to Nextcloud.
3. Place ./env file in same directory as the script you want to run, or edit running configuration to use .env file as environmental variables source (sample configurations for PyCharm are located in ./.idea\runConfigurations).
4. Use SGRFDatasetLoader.get_learning_files_nextcloud() function to load images from Nextcloud.

Sample usage:

import cv2
from scripts.loaders import SGRFDatasetLoader

# files = SGRFDatasetLoader.get_learning_files(limit=images_amount, limit_people=people_amount)
files = SGRFDatasetLoader.get_learning_files_nextcloud(limit_people=2, limit=100)
for image_file in files:
  image = cv2.imread(image_file[0])