Web App for Perspective Correction
Budget: $30 – $250 AUD
I want a lightweight, browser-based application that automatically straightens converging verticals in real-estate photos. Accuracy of correction is the top priority; I need the result to look as if the building was photographed with a perfectly level camera. ShiftN (shiftn.de) comes close, but often leaves residual tilt or unnatural stretching, so the new tool must be noticeably more reliable.
Core workflow
• User drags one or many images onto the page.
• The app detects vanishing points, computes the homography, and warps each photo so the main verticals are perfectly parallel.
• A side-by-side before/after viewer lets the user verify the fix.
• Corrected files are downloadable individually or in a zip when batch mode is used.
Key requirements
1. Accuracy first: the algorithm should handle wide-angle shots, slight rolls, and images with minimal vertical cues.
2. Batch processing and an interactive before-after comparison are mandatory.
3. Front-end must run in any modern browser; a minimal back-end is fine as long as uploads are secure and temporary.
4. Code must be clean and well-documented so I can extend it later.
I have no fixed stack preference, so feel free to propose OpenCV, TensorFlow, or any combination of JavaScript (e.g. WebAssembly-compiled OpenCV), Python (FastAPI, Flask), or other technologies you’re comfortable with. Please outline:
• The libraries and frameworks you’ll rely on.
• Your approach to vanishing-point detection and homography estimation (e.g., RANSAC line clustering, deep-learning refinement, etc.).
• Expected development timeline with major milestones.
• Fixed price or milestone-based quote.
Generic copy-paste replies will be ignored; I’m looking for a concise explanation of your method and why it will produce more accurate results than ShiftN.
Core workflow
• User drags one or many images onto the page.
• The app detects vanishing points, computes the homography, and warps each photo so the main verticals are perfectly parallel.
• A side-by-side before/after viewer lets the user verify the fix.
• Corrected files are downloadable individually or in a zip when batch mode is used.
Key requirements
1. Accuracy first: the algorithm should handle wide-angle shots, slight rolls, and images with minimal vertical cues.
2. Batch processing and an interactive before-after comparison are mandatory.
3. Front-end must run in any modern browser; a minimal back-end is fine as long as uploads are secure and temporary.
4. Code must be clean and well-documented so I can extend it later.
I have no fixed stack preference, so feel free to propose OpenCV, TensorFlow, or any combination of JavaScript (e.g. WebAssembly-compiled OpenCV), Python (FastAPI, Flask), or other technologies you’re comfortable with. Please outline:
• The libraries and frameworks you’ll rely on.
• Your approach to vanishing-point detection and homography estimation (e.g., RANSAC line clustering, deep-learning refinement, etc.).
• Expected development timeline with major milestones.
• Fixed price or milestone-based quote.
Generic copy-paste replies will be ignored; I’m looking for a concise explanation of your method and why it will produce more accurate results than ShiftN.