PyTorch to TensorFlow AI Conversion

1. Project Overview
We are looking for a freelancer who can convert an AI model from PyTorch to TensorFlow, based on the following paper:
Title: Learning Image-adaptive 3D Lookup Tables for High Performance Photo Enhancement in Real-time
Paper: https://arxiv.org/abs/2009.14468
Official Code (PyTorch): https://github.com/HuiZeng/Image-Adaptive-3DLUT
This paper presents a lightweight deep learning framework that learns image-adaptive 3D LUTs for real-time tone and color enhancement of high-resolution images.
2. Scope of Work
2.1 Reimplementation of the official PyTorch code in TensorFlow 2.x (with Keras), replicating the full functionality
Key components:
• Learnable 3D LUTs (33×33×33 grid with trilinear interpolation)
• Lightweight CNN (predicting LUT weights from downsampled input)
• Support for both paired and unpaired training modes
• Loss functions include: MSE, GAN loss, smoothness and monotonicity regularizations
Deliverable validation:
• Run training and inference using the same dataset (e.g., MIT-Adobe FiveK)
• Compare TensorFlow output images to PyTorch results
• Evaluate using PSNR, SSIM, and Delta E metrics
2.2 Support for PyTorch environment setup, debugging, and execution (tools and Q&A)

3. Final Deliverables
Fully implemented TensorFlow version of the AI model (including training and inference code), modularized with comments
Performance comparison against the original PyTorch version (same inputs, same performance)
Metrics and visual results for comparison
4. Timeline
Start date to be discussed
Target completion: within 2–3 weeks