v0.1.4

Update dependencies ort to 2.0.0.rc9.

Add new model config files for MegaDetectorV6-yolov9e

v0.1.3

Update dependencies ffmpeg-sidecar to 2.0, ort to 2.0.0.rc8, onnxruntime to 1.20.0.

Bug fixes:

• Fix potential panic caused by ffmpeg error

v0.1.2

Features:

• Add multiple device support. Example to run 1 workers on device 0 and 2 worker on device 1:$ md5rs -d 0 -w 1 -d 1 -w 2, run both on integrated GPU and discrete GPU: $ md5rs -d gpu -w 1 -d 0 -w 2.

Bug fixes:

• Fix resize_with_pad wrong returned dimension.

v0.1.1

Features:

• Switch default resize algorithm to nearest for faster preproces.
• Support custom model config file. Now --model option takes a .toml model config file path. And --imgsz option is removed. Check example model config file here. Note that the input and output shape sequence and precision of custom model still need to be the same as predefined models.
• Check execution provider availability only once and save it in epifo_{device}.json file to save initialization time. If you want to force recheck, delete the file.

v0.1.0

ci: :wrench: add aarch64 macos release

Prebuilt ONNX Runtime lib with OpenVINO, CUDA, TensorRT, DirectML support

ort-prebuilt

feat: :sparkles: add buffer size arg, detailed arg description