Project Overview
In this project, we aimed to develop a model for estimating edema from facial images. The project involved building a robust model with specific metrics and testing results, along with visualization outputs like Grad-CAM.
Key Details
- Evaluation Metrics: We used Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and correlation coefficients to evaluate the model’s performance.
- Model Improvement: Our target was to achieve a precision of +/- 2lbs. We explored various techniques such as adding additional layers, feature extraction, data augmentation, regularization, and hyperparameter tuning.
- Visualization: We implemented visualization methods like Grad-CAM to provide insights into the model’s decision-making process.
- Model Training: We trained the model multiple times, experimenting with different optimizers (Adam, SGD, Adagrad) and loss functions (MSE, Huber Loss, L1 Loss, Smooth L1 Loss). Huber Loss provided the best results.
- Performance: The final model showed significant improvement in accuracy and reliability, meeting the client’s requirements within the budget constraints.
Achievements
Conclusion
This project was a great success, demonstrating our ability to deliver high-quality results within a limited budget. The model’s performance and the insights gained from visualization techniques have paved the way for future improvements and applications.
Edema Estimation from Facial Images