HDRTVDM-onnx

This repo is a modified fork of AndreGuo/HDRTVDM, adapted for efficient ONNX inference on CPU-only environments. HDRTVDM is a deep learning model for converting LDR images into HDR outputs with BT.2020 precision and format flexibility.

🚀 What’s New in This Version

• ✅ Updated test.py for Torch-CPU compatibility (no GPU required)
• ✅ Added export_onnx.py for ONNX model export
• ✅ Includes 3 pre-converted ONNX files for direct use

🧪 Demo Comparison Pictures

Before/After demo: https://whyb.github.io/HDRTVDM-onnx/

You can evaluate the enhancement quality by comparing the original SDR pictures with the HDR outputs generated by HDRTVDM:

Before (0_SDR)	After (0_HDR)

Before (1_SDR)	After (1_HDR)

Before (2_SDR)	After (2_HDR)

Before (3_SDR)	After (3_HDR)

Before (4_SDR)	After (4_HDR)

Before (5_SDR)	After (5_HDR)

Before (6_SDR)	After (6_HDR)

All demo files are included in this repo.

🛠️ Usage

Run Inference with ONNX

Coming soon — inference with ONNXRuntime will be supported.

Export Custom ONNX Model

python method/export_onnx.py --output export/TriSegNet.onnx --height 1080 --width 1920