binary image classification (radiobuttons from scanned paper)
Budget: €30 – €250 EUR
## Description
I require a talented and dedicated professional who can deliver high-quality results for my binary image classification project.
From scanned paper questionnaires, I have around 270,000 unticket radiobuttons, 67,700 ticked radiobuttons and 10,529 anomalies.
In anomalies, it is unclear whether the fields were ticked, revoked or have other issues.
Three classes must be created: "ticked", "unticked" and "anomaly". The anomalies are then checked manually by a human.
## Design and Development
Design and develop a tailored model for binary image classification into the three classes.
Implement data preprocessing and augmentation techniques to improve model generalization.
Experiment with different architectures and hyperparameters to achieve optimal performance.
## Training and Evaluation
Train the model on provided image datasets (2 color)
Evaluate model performance using appropriate metrics and validation techniques.
Implement cross-validation to ensure robust performance across different subsets of the data.
## Regularization and Generalization
Implement regularization techniques (e.g., dropout, weight decay) to mitigate overfitting.
Utilize data augmentation strategies to enhance the diversity of the training data.
Apply techniques such as early stopping to prevent overtraining.
## Uncertainty Estimation
Incorporate Monte Carlo methods to estimate the uncertainty in model predictions.
Implement and analyze techniques like dropout during inference to provide confidence intervals for predictions.
## Optimization and Deployment
Optimize the model for computational efficiency and scalability.
Prepare the model for deployment, ensuring it is production-ready.
Document the entire process, including model architecture, training procedures, and performance evaluation.
## Requirements
Proven experience in developing and deploying models for image classification tasks.
Understanding of deep learning frameworks such as TensorFlow or Keras.
Proficiency in Python and relevant libraries (NumPy, OpenCV, scikit-learn).
Solid knowledge of techniques to handle overfitting, such as dropout, weight decay, and data augmentation.
Experience with Monte Carlo methods for uncertainty estimation in neural networks.
Familiarity with cross-validation and other validation techniques.
Excellent problem-solving skills and the ability to work independently.
## Preferred Qualifications
Previous experience with image classification tasks.
Experience with version control systems like Git.
Knowledge of deployment frameworks and tools (e.g., TensorFlow Serving, ONNX).
I require a talented and dedicated professional who can deliver high-quality results for my binary image classification project.
From scanned paper questionnaires, I have around 270,000 unticket radiobuttons, 67,700 ticked radiobuttons and 10,529 anomalies.
In anomalies, it is unclear whether the fields were ticked, revoked or have other issues.
Three classes must be created: "ticked", "unticked" and "anomaly". The anomalies are then checked manually by a human.
## Design and Development
Design and develop a tailored model for binary image classification into the three classes.
Implement data preprocessing and augmentation techniques to improve model generalization.
Experiment with different architectures and hyperparameters to achieve optimal performance.
## Training and Evaluation
Train the model on provided image datasets (2 color)
Evaluate model performance using appropriate metrics and validation techniques.
Implement cross-validation to ensure robust performance across different subsets of the data.
## Regularization and Generalization
Implement regularization techniques (e.g., dropout, weight decay) to mitigate overfitting.
Utilize data augmentation strategies to enhance the diversity of the training data.
Apply techniques such as early stopping to prevent overtraining.
## Uncertainty Estimation
Incorporate Monte Carlo methods to estimate the uncertainty in model predictions.
Implement and analyze techniques like dropout during inference to provide confidence intervals for predictions.
## Optimization and Deployment
Optimize the model for computational efficiency and scalability.
Prepare the model for deployment, ensuring it is production-ready.
Document the entire process, including model architecture, training procedures, and performance evaluation.
## Requirements
Proven experience in developing and deploying models for image classification tasks.
Understanding of deep learning frameworks such as TensorFlow or Keras.
Proficiency in Python and relevant libraries (NumPy, OpenCV, scikit-learn).
Solid knowledge of techniques to handle overfitting, such as dropout, weight decay, and data augmentation.
Experience with Monte Carlo methods for uncertainty estimation in neural networks.
Familiarity with cross-validation and other validation techniques.
Excellent problem-solving skills and the ability to work independently.
## Preferred Qualifications
Previous experience with image classification tasks.
Experience with version control systems like Git.
Knowledge of deployment frameworks and tools (e.g., TensorFlow Serving, ONNX).