MATLAB CNN Fingerprint Reconstruction Model
Budget: ₹600 – ₹1,500 INR
I need a clean, fully reproducible implementation of a convolutional neural network that can reconstruct partially missing fingerprints. The workflow starts by masking out regions of each print, forcing the network to learn how to in-paint those gaps. Because this is for research, every step—from data creation to final evaluation—has to be transparent, well commented, and easy to repeat on another machine.
Core requirements
• Codebase in MATLAB is my first choice. If a critical library isn’t available, a fallback in Python under Anaconda (TensorFlow or PyTorch) is acceptable, but explain the reason for the switch up front.
• A custom fingerprint dataset is expected. It must cover a wide variety of fingerprint patterns so the model generalises well; I’m not looking for off-the-shelf public sets. Include scripts that either synthesise or curate this data and document the exact procedure so I can regenerate it from scratch later.
• Automatic mask generation with controllable size, position, and shape parameters.
• End-to-end training and testing scripts that output clear metrics (e.g., PSNR, SSIM, reconstruction accuracy) along with visual before/after examples. Cross-verify results by running the trained model on the held-out test split in a separate session.
• Save and ship the trained model weights plus any configuration files.
• Inline comments and a brief README that walks through installation, dataset generation, training, testing, and reproducing the reported numbers.
Acceptance criteria
1. I can run a single command to rebuild the dataset, train the network, and reproduce your reported metrics.
2. No errors or missing dependencies when the project is executed in a fresh environment.
3. All plots, logs, and numeric results match (within normal random-seed variation) the figures you supply.
Payment is released only after these criteria are met and every requested script, model, and dataset component is delivered. If you are confident you can satisfy the reproducibility bar, I look forward to seeing your approach.
Core requirements
• Codebase in MATLAB is my first choice. If a critical library isn’t available, a fallback in Python under Anaconda (TensorFlow or PyTorch) is acceptable, but explain the reason for the switch up front.
• A custom fingerprint dataset is expected. It must cover a wide variety of fingerprint patterns so the model generalises well; I’m not looking for off-the-shelf public sets. Include scripts that either synthesise or curate this data and document the exact procedure so I can regenerate it from scratch later.
• Automatic mask generation with controllable size, position, and shape parameters.
• End-to-end training and testing scripts that output clear metrics (e.g., PSNR, SSIM, reconstruction accuracy) along with visual before/after examples. Cross-verify results by running the trained model on the held-out test split in a separate session.
• Save and ship the trained model weights plus any configuration files.
• Inline comments and a brief README that walks through installation, dataset generation, training, testing, and reproducing the reported numbers.
Acceptance criteria
1. I can run a single command to rebuild the dataset, train the network, and reproduce your reported metrics.
2. No errors or missing dependencies when the project is executed in a fresh environment.
3. All plots, logs, and numeric results match (within normal random-seed variation) the figures you supply.
Payment is released only after these criteria are met and every requested script, model, and dataset component is delivered. If you are confident you can satisfy the reproducibility bar, I look forward to seeing your approach.