Method for creating a deep learning-based model and device for implementing the model created by said method
A method comprising: providing a plurality of samples to form a dataset; providing a neural network; training the neural network with said dataset; evaluating the model, adjusting the model until an accuracy threshold is exceeded; wherein to produce the samples of the training sample set of the dataset: a stage with an object on it is provided, a background of a first color and a light source constituting a first combination is arranged, at least one image is captured, consecutively different combinations are arranged and at least one image is captured, the images are cropped to remove the part of the image that does not contain an object, each of the images is labelled with a label identifying the object category, the resulting labelled images forming part of the plurality of samples to form the dataset.
1 . A method for creating a deep learning based model to solve a question for a pre-determined category of objects, comprising:
I) providing a plurality of samples to form a dataset, store such dataset and partition such dataset into three sample groups: training sample group, evaluation sample group, and prediction sample group,
II) determining an architecture of a neural network and providing a corresponding neural network,
III) providing the training sample group of samples from the dataset to the neural network, and thereby creating a model,
IV) causing the model to learn from the training sample group of samples in the dataset by training the model through detecting features and patterns,
V) saving the trained model,
VI) providing the group of evaluation samples from the dataset to the trained model to output a result for each sample, and evaluate the accuracy of the results,
VII) if the accuracy of the results according to the evaluation of step VI does not exceed a predetermined threshold, adjusting the trained model and repeating steps IV to VII; if the accuracy of the results according to the evaluation of step VI exceeds a predetermined accuracy threshold, validating the model;
wherein the samples of the training sample group of the dataset are produced by the following procedure:
i) a stage is provided, in which an object from the pre-determined object category is placed,
ii) a background of a first color is placed on the stage and a light source of a first wavelength illuminates the stage-object set, which is a first combination of stage background color and illumination wavelength of the stage-object set,
iii) at least one image of the scene-object set is captured,
iv) one or more different combinations of stage background color and illumination wavelength of the stage-object set are arranged consecutively and at least one image of the stage-object set is captured with each combination,
V the captured images are processed by an object detection component and cropped in such a way that the parts of the images that do not contain the object are removed,
vi) each of the processed images are tagged with a label that identifies the category of the object, the resulting labelled images being part of the plurality of samples to form the dataset.
2 . The method according to claim 1 , further comprising:
VIII) feeding a sample from the set of prediction samples to the model, from among the prediction samples not fed to the model, for the model to output a result, and verifying that the result is correct,
IX) if there are no samples in the prediction sample set that have not been fed to the model, confirming the validity of the model, or
when there are samples in the prediction sample pool that have not been fed to the model:
if the verification of the result in step VIII indicates that the result is not correct, adjusting the model and repeating steps IV to IX,
if the verification of the result in step VIII indicates that the result is correct, repeating steps VIII and IX.
3 . The method according to claim 1 , wherein, if a plurality of images are captured with the same combination of stage background color and wavelength of illumination of the stage-object, at least two such images may be captured with the object in positions different from each other.
4 . The method according to claim 1 , further comprising the following process steps:
vii) the object on the stage is removed from the stage, another object from the predetermined object category is placed on the stage, and steps ii to vii are repeated.
5 . The method according to claim 1 , wherein in process steps iii and iv, a video is made and frames from which images are to be captured are extracted from the video.
6 . The method according to claim 1 , wherein the object placed on the stage is the object itself or a support with a graphical representation of the object or a three-dimensional representation of the object.
7 . A device for implementing a deep learning based model created in accordance with the method of claim 1 .
8 . A support containing a deep learning based model created according to the method of claim 1 .