Upload README.md

README.md

@@ -1,5 +1,5 @@
 ---
-title: UVDoc Document Unwarping
+title: UVDoc Document Unwarping (High-Resolution)
 emoji: 📄
 colorFrom: blue
 colorTo: green
@@ -10,22 +10,18 @@ pinned: false
 license: apache-2.0
 ---
 
-# UVDoc Document Unwarping
+# UVDoc Document Unwarping (High-Resolution)
 
-Interactive demo of **UVDoc**, a neural grid-based document unwarping model.
+Interactive demo of **UVDoc** using the ONNX grid-output model for **native-resolution** document unwarping.
 
-Upload a photo of a warped, curved, or perspective-distorted document, and the model will geometrically rectify it.
+## Why this version?
 
-## Model
-
-- [PaddlePaddle/UVDoc_safetensors](https://huggingface.co/PaddlePaddle/UVDoc_safetensors) — Transformers-compatible safetensors checkpoint.
+- The original PaddlePaddle/UVDoc_safetensors checkpoint outputs a fixed low-res rectified image.
+- This space uses the ONNX grid model (`fredcallagan/uvdoc-grid-onnx`) which outputs a coordinate grid. That grid is then applied to your **original full-resolution image** via `cv2.remap`, preserving detail.
 
-## How it works
+## Model
 
-1. The image is resized and normalized via `UVDocImageProcessor`.
-2. `UVDocModel` predicts spatial transformation parameters.
-3. `post_process_document_rectification` warps the image back to a flat rectangle.
-4. The rectified image is returned.
+- [fredcallagan/uvdoc-grid-onnx](https://huggingface.co/fredcallagan/uvdoc-grid-onnx)
 
 ## Paper