Release notes
# Release 1.3.3
We’re excited to announce the release of **v1.3.3**! This update brings significant improvements and expanded support, especially for Windows GPU users and ONNX Runtime providers. Here’s what’s new and improved since 1.3.2:
---
## 🚀 Major Improvements
### Windows GPU Support Returns!
- **Windows GPU builds are back:** Now you can easily build and use the GPU version on Windows. Build artifacts are named more clearly (`-gpu` suffix) for easier downloads and usage.
### Enhanced ONNX Runtime Support
- **CUDA & ROCm Providers:** Linux users can now select ROCm in addition to CUDA for GPU inference! The UI shows these options only when supported by your build and ONNX Runtime setup.
- **Cleaner configuration:** Default CMake options now better support both system and bundled ONNX Runtime, with clear platform presets to avoid confusion.
### TensorRT Changes
- **Reworked TensorRT support:** Added, removed, and refined support for the ONNX Runtime TensorRT execution provider on Windows. Note: This option’s visibility may vary depending on your platform and build configuration.
## 🖌️ Usability Enhancements
- **Mask Expansion Tool:** A new slider under advanced settings allows you to expand the safe area around detected persons. Fine-tune your segmentation results with ease!
- **Localized ROCm Option:** The “ROCm” GPU option now appears properly in all supported languages, offering improved clarity for non-English users.
## 🛠️ Under-the-Hood & Developer Experience
- **Improved CI/CD workflows:** Automated artifact renaming and simplified DLL/PDB installation on Windows.
- **Refactored build logic:** Better macro names and configuration flags for ONNX Runtime GPU providers, resulting in more robust and maintainable code.
- **Documentation and Help UI:** Enhanced installation instructions and a more responsive, English-friendly help and chatbot interface (for those using the interactive chat help).
---
## ⚠️ Notes for Upgraders
- **Artifact names:** GPU-enabled Windows builds are now labeled with `-gpu` for crystal clear distinction.
- **TensorRT:** Support may be present or absent depending on your configuration—refer to documentation or your build options.
- **Recommended:** If you’re updating custom CMake presets or localized files, check for new or renamed macro/configuration entries!
---
Thank you to all contributors and users for your feedback and support! We're committed to making your experience better on every platform. Try out the new features and let us know what you think!
Happy building! 🚀