Real-ESRGAN NCNN Image Upscaler

🔧Quick Start

1. Extract Zip and run RealESRGAN-NCNN.exe to run the application.
2. Check out Usage below.

Go through code.

1. Extract this ZIP file to any folder
2. Download Real-ESRGAN NCNN models and place them in the models folder
3. Download Real-ESRGAN NCNN executable and place it in the bin folder
4. If you don't want to download manually then run download_dependencies.py file directly.
5. Run build.py file for generate RealESRGAN-NCNN.exe.
6. you have to replace dist_final models and bin to download_dependencies.py models and bin.
7. Double-click RealESRGAN-NCNN.exe to run the application

👀Required Files Structure:

RealESRGAN-NCNN/
├── RealESRGAN-NCNN.exe          # Main application
├── bin/
│   └── realesrgan-ncnn-vulkan.exe   # NCNN executable
├── models/
│   ├── realesrgan-x4plus.param
│   ├── realesrgan-x4plus.bin
│   ├── realesrgan-x4plus-anime.param
│   ├── realesrgan-x4plus-anime.bin
│   └── ...
└── _internal/                   # Application dependencies (auto-generated)

🏁 Result

Click Me

⚡Usage:

1. Click Load Image to select an image
2. Choose model and scale settings
3. Click Process Image to upscale
4. Click Save Result to save the upscaled image

Supported Formats:

• Input: PNG, JPG, JPEG, BMP, TIF, TIFF, WEBP
• Output: JPG, PNG, WEBP

💻System Performance:

Windows	Processor	RAM	GPU	Toal Time(s)	Perfomance
10	I3-I5 (<6 Gen)	4/4+	Default	1500-2100	Too Slow
10	I7	8+	Default	600	Slow
10	I7	8+	GTX or AMD GPU	300	Quite Good
11	I5 (>6 Gen)	8/8+	Default	120-300	Good
11	I7	8+	Default	120-250	Great
11	I7	8+	GTX or AMD GPU	100-150	Impressive
11	I7	16+	RTX or AMD GPU	15-30	Execellent

Suggestion

• Windows 11 (64-bit)
• Processor I5 12th gen beat I7 otherwise I7 latest
• Compatible GPU: RTX or amd rx best if have GTX then its fine
• RAM: 8GB+ RAM
• 1GB+ free disk space (For storing output images)

🚧Troubleshooting:

• If the app doesn't start, ensure all files are extracted properly
• If processing fails, check that the NCNN executable is in the 'bin' folder
• For GPU acceleration, ensure compatible drivers are installed

💪Performance over CPU / GPU

• CPU: Takes around 15-20 min for per image.
• GPU: Depend on GPU in my case I have NVIDIA RTX 4060 takes around 20-30 sec per img.

📝Note

• Always set Scale settings on 4x. if set in 2x raise error because model param's can only compute for 4x.
• Always prefer less than 2k image dimentions, check h * w before processing otherwise error occur for exceeding the image pixels. You can initally decrease their pixels because after processing you get 4k level image.

🚩Updated Build Releases

• Consider latest releases at https://github.com/radadiavasu/Real-ESRGAN-NCNN/releases/ section.