batch inferece for retinaface

Hi,
It seems that detection model, i.e. retinaface, only supports single image inference. Since the model provided is in ONNX format. How can I re-convert pth model to ONNX model since pre-trained pth weight is not provided?
Any help will be appreciate!

Start by reviewing the RetinaFace ONNX inference path and the repository's model/export documentation. Determine whether the requested batch inference and conversion from pretrained weights are supported by the available artifacts, then define the expected batch behavior and validation needed to consider the issue done.

Written by the indexing model from the issue text.