We are building an AI-powered sky replacement system and need an experienced developer to prepare a training dataset and implement an optimized image processing flow. 1. Dataset Creation 1.1. You will receive approximately 2000 unedited DSLR images. 1.2. Create accurate sky masks for each image (input + mask pair). 1.3. Masks must be precise around trees, hair, and reflections, and suitable for AI training. 2. Processing Flow Development 2.1. Upload Original (~5 MB) images. 2.2. Auto Compress/Resize while maintaining high visual quality for fast web preview. 2.3. Sky Detection – if sky area is less than 5%, skip the sky replacement process. 2.4. AI Sky Replacement – replace detected sky with realistic blending and sharp masking. 2.5. Save Optimized Version for web display or preview. 2.6. On user download, apply AI Super Resolution / Enhancer to upscale to approximately 5–6 MB and restore full detail and clarity. 3. Goal A complete smart pipeline: > Large image → compress → detect sky → skip or replace → store optimized → upscale when downloaded. 4. Deliverables 4.1. 2000-image sky mask dataset (input + mask pairs). 4.2. Working Python-based AI sky replacement pipeline, preferably using U²-Net, MODNet, or a similar architecture. 4.3. Quality-tested blending and enhancement outputs.