# Request: RF-DETR (Roboflow) Neural Compiler Extension

**URL:** <https://community.memryx.com/t/request-rf-detr-roboflow-neural-compiler-extension/226>\
**Category:** Technical Support\
**Created:** [July 10, 2026, 8:31am UTC](https://community.memryx.com/t/request-rf-detr-roboflow-neural-compiler-extension/226 "2026-07-10T08:31:35Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Miguel](https://avatars.discourse-cdn.com/v4/letter/m/439d5e/32.png) [@Miguel](https://community.memryx.com/u/Miguel)\
**Post date:** [July 10, 2026, 8:31am UTC](https://community.memryx.com/t/request-rf-detr-roboflow-neural-compiler-extension/226/1 "2026-07-10T08:31:35Z")

</div>

We have an RF-DETR face detection model (exported to ONNX) we’d like to compile to DFP on SDK 2.2. The compiler fails on GridSample, TopK, GatherElements, Where (ops from the multi-scale deformable attention decoder). Given you already have an RT-DETRv2 subgraph handler (architecturally very similar), would a custom NCE extension be possible?

---

<div class="post-metadata">

**Author:** ![sureshvairamuthu](https://avatars.discourse-cdn.com/v4/letter/s/e99b99/32.png) [@sureshvairamuthu](https://community.memryx.com/u/sureshvairamuthu)\
**Post date:** [July 16, 2026, 2:50pm UTC](https://community.memryx.com/t/request-rf-detr-roboflow-neural-compiler-extension/226/2 "2026-07-16T14:50:17Z")

</div>

Hello @Miguel,

Thank you for reaching out and working with MX3 product.

A custom Neural Compiler Extension for RF-DETR is technically feasible. However, on the current-generation MX3 platform, support for some transformer-specific operators and graph patterns is still limited, particularly in deformable-attention decoder blocks.

The RT-DETR extension you referenced was developed primarily as an internal engineering prototype to evaluate this class of architecture. It accelerates only a portion of the RT-DETR graph rather than providing complete end-to-end model support, so it is not currently offered as a general-purpose public extension.

We are continuing to expand transformer support as part of our next-generation product work.

For face detection on MX3 today, a YOLO-based detector is likely to be the most straightforward and effective option. In our experience, transformer-based detectors can add significant architectural complexity for this particular use case without necessarily providing a final deployment advantage.

You may find this real-time multi-face recognition pipeline useful as a starting point:

> **[MemryX\_eXamples/video\_inference/realtime\_multiface\_recognition at release ·...](https://github.com/memryx/MemryX_eXamples/tree/release/video_inference/realtime_multiface_recognition)**
>
> release/video\_inference/realtime\_multiface\_recognition