[Bug] How to obtain RTMO or YOLOX-Pose with 960 input size (ONNX export)?

Prerequisite

• I have searched Issues and Discussions but cannot get the expected help.
• The bug has not been fixed in the latest version(https://github.com/open-mmlab/mmpose).

Environment

Background I'm deploying multi-person pose estimation on high-resolution RTSP streams (2048×1536) where many people appear small/far from the camera. With the default 640×640 input, small/distant people are frequently missed. I'd like to try a 960×960 input to see whether it improves detection of small people.

What I've found so far The official RTMO ONNX models (e.g. [rtmo-m_16xb16-600e_body7-640x640]have their input H/W hard-coded to 640×640 (['batch', 3, 640, 640]) — only the batch dimension is dynamic. So I cannot simply feed 960 input; ONNX Runtime raises a shape error.

For YOLOX-Pose, the project README explicitly states:

"We have only trained models with an input size of 640, as we couldn't replicate the performance enhancement mentioned in the paper when increasing the input size from 640 to 960."

So no 960 weights are published there either.

My questions Is there any released RTMO or YOLOX-Pose checkpoint / config trained (or usable) at 960×960 input? If not officially, has anyone in the community succeeded?

What is the correct way to export an ONNX model with 960×960 input from the existing 640-trained weights (via MMDeploy or tools/)? Specifically:

Which deploy config field controls input_shape, and is it safe to just change 640 → 960? Are the post-processing steps (keypoint decoding / coordinate scaling) input-size-agnostic, or do they assume 640? Is it possible / recommended to export a dynamic H/W ONNX (one model that accepts both 640 and 960), or do the reshape/concat nodes in RTMO make this unreliable?

Accuracy expectation: For a 640-trained RTMO/YOLOX-Pose model, is running inference at 960 (resolution mismatch) expected to help small-person detection, or does it typically degrade overall accuracy? Any experience appreciated.

Environment onnxruntime-gpu 1.22, PyTorch 2.5.1+cu121, Windows Model: RTMO-m (body7, 640×640) Thank you very much for any guidance!

Reproduces the problem - code sample

.

Reproduces the problem - command or script

.

Reproduces the problem - error message

.

Additional information

No response