BSRGANx4_fp16

BSRGANx2_fp16

BSRGANx4_fp16 is an ONNX export of a BSRGAN-based super-resolution model for 4× image upscaling.

Model Description

This repository provides a float16 ONNX version of a BSRGAN super-resolution model intended for deployment in ONNX Runtime based applications and custom upscaling pipelines.

File

• BSRGANx4_fp16.onnx

Intended Use

This model is intended for:

• 4× image upscaling
• video frame enhancement
• ONNX Runtime inference workflows
• integration into desktop and batch processing pipelines

Input / Output

• Input: low-resolution image tensor prepared by the host application
• Output: 4× upscaled image

Exact preprocessing, color handling, and tensor formatting depend on the application using the model.

Usage Notes

• Designed for ONNX Runtime
• FP16 export for improved performance on supported hardware
• Can be used with CUDA, TensorRT, DirectML, or CPU execution providers depending on environment support

Source / Attribution

This repository contains an ONNX-packaged version of a BSRGAN-based super-resolution model.

The original BSRGAN project and its upstream licensing are associated with the cszn/BSRGAN repository.

Limitations

• GAN-based upscaling may introduce hallucinated or stylized detail
• Output quality depends on source resolution, compression, and host pipeline implementation
• Not intended for measurement-accurate reconstruction or scientific imaging use

License

Apache License 2.0