Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,64 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: bsd-3-clause
|
| 3 |
-
--
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: bsd-3-clause
|
| 3 |
+
pipeline_tag: image-to-image
|
| 4 |
+
library_name: onnxruntime
|
| 5 |
+
tags:
|
| 6 |
+
- onnx
|
| 7 |
+
- super-resolution
|
| 8 |
+
- upscaling
|
| 9 |
+
- real-esrgan
|
| 10 |
+
- image-processing
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
# RealESRGANx4_fp16
|
| 14 |
+
|
| 15 |
+
RealESRGANx4_fp16 is an ONNX export of a Real-ESRGAN-based super-resolution model for 4× image upscaling.
|
| 16 |
+
|
| 17 |
+
## Model Description
|
| 18 |
+
|
| 19 |
+
This repository provides a float16 ONNX version of a Real-ESRGAN super-resolution model designed to enhance image detail and visual quality across a variety of content.
|
| 20 |
+
|
| 21 |
+
This variant follows the standard Real-ESRGAN approach and may produce sharper, more stylized results compared to more balanced variants.
|
| 22 |
+
|
| 23 |
+
## File
|
| 24 |
+
|
| 25 |
+
- `RealESRGANx4_fp16.onnx`
|
| 26 |
+
|
| 27 |
+
## Intended Use
|
| 28 |
+
|
| 29 |
+
This model is intended for:
|
| 30 |
+
|
| 31 |
+
- 4× image upscaling
|
| 32 |
+
- video frame enhancement
|
| 33 |
+
- detail enhancement in images and video pipelines
|
| 34 |
+
- ONNX Runtime based applications
|
| 35 |
+
|
| 36 |
+
## Input / Output
|
| 37 |
+
|
| 38 |
+
- Input: low-resolution image tensor
|
| 39 |
+
- Output: 4× upscaled image
|
| 40 |
+
|
| 41 |
+
Exact preprocessing and normalization depend on the host application.
|
| 42 |
+
|
| 43 |
+
## Usage Notes
|
| 44 |
+
|
| 45 |
+
- Emphasizes sharper and more detailed output
|
| 46 |
+
- FP16 export for improved performance on supported hardware
|
| 47 |
+
- Compatible with ONNX Runtime (CUDA, TensorRT, DirectML, CPU)
|
| 48 |
+
- May produce more aggressive detail enhancement than general-purpose models
|
| 49 |
+
|
| 50 |
+
## Source / Attribution
|
| 51 |
+
|
| 52 |
+
This repository contains an ONNX-packaged version of a Real-ESRGAN-based model.
|
| 53 |
+
|
| 54 |
+
The upstream Real-ESRGAN project is licensed under the BSD 3-Clause License.
|
| 55 |
+
|
| 56 |
+
## Limitations
|
| 57 |
+
|
| 58 |
+
- GAN-based models may introduce artificial or exaggerated detail
|
| 59 |
+
- May oversharpen certain textures or edges
|
| 60 |
+
- Output quality depends on input resolution and preprocessing
|
| 61 |
+
|
| 62 |
+
## License
|
| 63 |
+
|
| 64 |
+
BSD 3-Clause
|