Real-ESRGAN
Real-ESRGAN is a super-resolution algorithm developed by Tencent ARC Lab designed to restore real-world complex degraded images. It improves traditional GAN training methods by using second-order degradation models to simulate blur, noise, and compression artifacts in real images, enabling natural reconstruction of low-quality images.
- Key Features: Supports extremely high-quality image detail enhancement and artifact removal, significantly improving low-resolution image clarity while restoring texture quality. Widely used in old photo restoration, video enhancement, anime upscaling, and security image analysis.
- Version Note: This case uses the Real-ESRGAN_x4plus model. As the most generalized version in this family, it is specifically optimized for various unknown degradations in real-world scenarios. While maintaining the classic RRDB architecture, it achieves excellent balance between image clarity and visual realism through deeper feature extraction capabilities, making it the preferred solution for general image upscaling and restoration tasks.
Environment Setup
Configure the required environment in advance.
Quick Start
Download Model Files
O6 / O6N
cd ai_model_hub_25_Q3/models/ComputeVision/Super_Resolution/onnx_real_esrgan
wget https://www.modelscope.cn/models/cix/ai_model_hub_25_Q3/resolve/master/models/ComputeVision/Super_Resolution/onnx_real_esrgan/real_esrgan.cix
Model Testing
info
Activate the virtual environment before running!
O6 / O6N
python3 inference_npu.py
Complete Conversion Workflow
Download Model Files
Linux PC
cd ai_model_hub_25_Q3/models/ComputeVision/Super_Resolution/onnx_real_esrgan/model
wget https://www.modelscope.cn/models/cix/ai_model_hub_25_Q3/resolve/master/models/ComputeVision/Super_Resolution/onnx_real_esrgan/model/realesrgan-x4.onnx
wget https://www.modelscope.cn/models/cix/ai_model_hub_25_Q3/resolve/master/models/ComputeVision/Super_Resolution/onnx_real_esrgan/model/RealESRGAN_x4plus.pth
Project Structure
├── cfg
├── datasets
├── inference_npu.py
├── inference_onnx.py
├── model
├── pytorch2onnx_x4.py
├── README.md
├── real_esrgan.cix
└── test_data
Perform Model Quantization and Conversion
Linux PC
cd ..
cixbuild cfg/onnx_realesrganbuild.cfg
Push to Board
After completing the model conversion, push the cix model file to the board.
Test Host Inference
Run Inference Script
Linux PC
python3 inference_onnx.py
Model Inference Results
Deploy to NPU
Run Inference Script
O6 / O6N
python3 inference_npu.py
Model Runtime Results
O6 / O6N
$ python3 inference_npu.py
npu: noe_init_context success
npu: noe_load_graph success
Input tensor count is 1.
Output tensor count is 1.
npu: noe_create_job success
npu: noe_clean_job success
npu: noe_unload_graph success
npu: noe_deinit_context success
