Optical Character Recognition (OCR) Model

DeepLobe's OCR (Optical Character Recognition) model is designed to extract text from images or scanned documents, enabling users to convert physical or digital text into editable and searchable formats. OCR technology plays a vital role in various industries, including document processing, data entry, archival systems, and more.

The OCR model in DeepLobe leverages advanced machine learning algorithms to accurately detect and recognize text within images. It can handle a wide range of font styles, sizes, and orientations, making it versatile for different types of documents. Whether you need to extract text from printed documents, handwritten notes, or digital images, DeepLobe's OCR model provides reliable and efficient results.

By analyzing the visual patterns and features of text, the OCR model segments and recognizes individual characters, words, and sentences within an image. It is trained on large datasets to ensure high accuracy and robustness in text extraction tasks.

DeepLobe's no-code AI platform empowers users to create custom OCR models tailored to their specific needs. Whether you want to extract text from invoices, receipts, business cards, or any other document type, DeepLobe offers a user-friendly interface and powerful tools to train and deploy OCR models without the need for coding.

Next, we will guide you through the steps and workflow of creating a custom OCR model using DeepLobe's intuitive interface and powerful features.

Add Training Data

You can add train data in multiple ways. 

Upload Images

You can upload images to DeepLobe to train your OCR model. To upload images, click on the "Browse files" button on the project creation screen. Locate and select the images from your local storage.Ensure that your images are in JPG, PNG, or JPEG format and meet the following requirements like minimum of 50 files, each file size should not exceed 1 MB

Import from Data Sources

You can also import images from data sources like OneDrive and Google Drive.

Add Data using JSON files

Additionally, you have the option to upload annotated data in .json format. If you have already annotated your data externally, you can upload the annotations to DeepLobe. This allows you to utilize pre-existing annotations and speed up the training process. To add data using JSON files:

• Click on the "Upload JSON" button on the project creation screen.
• Select the JSON file containing the annotations.
• DeepLobe will use this annotated data to train your OCR model.

It's important to have accurate and comprehensive annotations for training your custom model effectively. Well-annotated data helps the model learn and identify objects accurately in images or videos.

Annotate Training Data

To annotate your dataset for the custom OCR (Optical Character Recognition) model in DeepLobe, you can follow a straightforward annotation process. Here are the steps:

Upload and Prepare Your Dataset

Ensure that you have a minimum of 50 files in your dataset containing images or scanned documents. Once uploaded, click on "Continue" to proceed to the labeling tool. DeepLobe's intuitive annotation tool allows you to manually label and annotate text within the images. Here's how:

1. An image or document from your dataset will be displayed in the labeling tool.
2. Select a text region by clicking and dragging the cursor to form a bounding box around the text. The box should encapsulate the entire text or paragraph accurately.
3. Assign a label or class to each text region by entering the corresponding text in the provided field.
4. For example, if the text region contains a person's name, enter the name as the label.
   Repeat this process for each text region in the image or document.

Review and Edit Annotations

After annotating a text region, you can review and edit the annotations as needed. You can adjust the position and size of the bounding boxes, modify the assigned text labels, or delete incorrect annotations. It is important to ensure that the annotations accurately represent the text within the image or document.

Annotate the Entire Dataset

Repeat the annotation process for each image or document in your dataset. Take your time to ensure accurate and comprehensive annotations for each text region. The quality of annotations directly impacts the performance of your OCR model.

Once you have completed annotating all the images or documents in your dataset, the tool automatically saves your annotations for the next steps of training a custom model.

If you prefer to have a professional annotator assist you, you can click on "Contact Us" and select "Hire Annotation Expert" to request this service.

Accurate and detailed annotations are crucial for training your OCR model effectively. They help the model learn to recognize and extract text accurately from various sources, improving the overall performance of your OCR applications.

Data Health

In the custom OCR model flow, DeepLobe provides a Data Health Check stage to evaluate the quality and balance of your training and validation data sets. This stage allows you to assess the health of your data and take necessary actions to improve the precision and performance of your models.

DeepLobe offers powerful analysis tools, such as Class Balance, to assist you in identifying potential gaps or imbalances within your data. These tools highlight classes with a minority of data, making it easier for you to identify areas where additional data is needed to improve the model's performance.

To improve the health of your data and enhance the accuracy of your model, consider the following steps:

Add More Data

If the analysis reveals that certain classes have a limited number of samples, it is advisable to add more data for those classes. Collecting additional samples will provide a more diverse and representative training set, enabling the model to learn effectively.

Evenly Distribute Training Data

Ensure that your training data is evenly distributed across different classes. Imbalanced training data, where certain classes have significantly more samples than others, can lead to biased models. By evenly distributing the training data, you create a balanced learning environment that promotes fair and accurate predictions for all classes.
By following these steps, you can improve the quality and health of your data, leading to better model performance and more reliable object detection results.

DeepLobe's Data Health Check stage empowers you to evaluate and optimize your training and validation data sets, ultimately enhancing the effectiveness and accuracy of your custom object detection models.

Train Model

Once the data health of your custom OCR model is verified and deemed satisfactory, you can proceed to train your model by clicking on the "Train" button. The training process involves using the labeled data to teach the model how to identify and classify objects in images accurately.

Upon initiating the training process, DeepLobe will commence training your model using the provided data. The training time may vary depending on the size of the dataset and the complexity of the text labels being extracted. Once the training is completed, you will receive an email notification indicating that your model has been trained and is ready for use.

During the training phase, DeepLobe's advanced algorithms analyze the data and optimize the model's parameters to improve its accuracy and performance. The model learns to recognize patterns, features, and characteristics of the objects within the training data, enabling it to make accurate predictions on new, unseen images.

It's important to note that the training process may require multiple iterations and adjustments to fine-tune the model's performance. DeepLobe's training algorithm optimizes the model based on feedback and evaluation metrics, such as precision, recall, and F1 score, to ensure optimal results.

Once you receive the notification that your model has been trained and is ready, you can proceed to the next stage, which involves evaluating and testing the model's performance using the validation and test datasets.

Evaluate and Test Model

Once your OCR model has been trained, it is important to evaluate its performance and test its accuracy. DeepLobe provides tools to help you with this process.

To test the model, you can upload any image of your choice and click on the "Test" button. The model will analyze the image and accurately detect the objects within it. It will then provide the correct labels for each detected object. This allows you to assess the model's performance and verify if it meets your expectations in terms of accuracy and object segmentation.

Retrain Model

If you find that the model's accuracy is not satisfactory, DeepLobe offers a retraining capability to help improve the model's performance. By adding more training data to the model, you can enhance its accuracy and increase its ability to correctly detect and segment objects. Simply upload your updated or additional data, and DeepLobe will take care of the retraining process. This makes it easy to iterate and improve your model with just a few clicks.

To initiate the retraining process, you can click on the "Retrain" button present on the results screen after testing the model. Additionally, you can navigate to the "My Models" section in the left panel of the DeepLobe platform. From there, you can select the specific model you want to retrain and click on the three-vertical dots corresponding to that model. Then, choose the "Retrain" option to start the retraining process.

By incorporating more data and retraining the model, you can achieve greater accuracy and performance in your OCR tasks. DeepLobe simplifies the retraining process, allowing you to optimize your model's performance with ease.

For more detailed guidance on evaluating model metrics, testing your model, and retraining, you can refer to the DeepLobe documentation or reach out to our support team for further assistance. We are here to help you achieve the best results with your custom instance segmentation models.

Model Integration

To integrate your custom Semantic Segmentation model into your applications or systems, you will need the model API endpoint and the corresponding API key. Follow these steps to obtain the API endpoint for your object detection model:

• Navigate to the "My Models" section in the DeepLobe platform.
• Locate your Semantic Segmentation model from the list of models.
• Look for the API endpoint associated with your model. It will be displayed in the platform, usually in the format of a URL or an endpoint address.
• Click on the copy icon located next to the API endpoint to copy it to your clipboard.

Once you have copied the API endpoint, you can use it to make API requests and integrate your object detection model into your applications or systems.

Note: The API endpoint is specific to your Semantic Segmentation model and allows you to communicate with the DeepLobe platform to utilize the model's capabilities.

For more information on how to use the API endpoint and integrate the model into your applications, you can refer to the DeepLobe API documentation or consult the relevant documentation for your programming language or framework.

API Key: To generate an API key for your DeepLobe account, follow these steps:

• Go to the "APIs" section in the DeepLobe platform.
• Look for the API key associated with your account.
• Click on the copy icon located next to the API key to copy it to your clipboard.
• Once you have obtained the API key, you can use it to authenticate and authorize your API requests to the DeepLobe platform.

For more information, go through our API documentation.