RT-DETRv2-S-OpenVINO

RT-DETRv2-S for Intel NPUs, GPUs and CPUs: rtdetr_graphA_fp16 and rtdetr_graphB_fp16 from litert-community/RT-DETRv2-S-LiteRT as ONNX (tf2onnx 1.17.0, opset 17), OpenVINO IR with fp32 weights, and precompiled Intel NPU blobs. tf2onnx 1.17.0 drops the bias of 13 FULLY_CONNECTED layers of rtdetr_graphA_fp16 and rtdetr_graphB_fp16; these files were converted with a one-line fix of that handler (convert/patch_tf2onnx.py in the GitHub repo), and their outputs match the .tflite. On the NPU of a Core Ultra 9 386H the blob runs in 12.07 ms (rtdetr_graphA_fp16) and 86.90 ms (rtdetr_graphB_fp16) per run, numpy input to numpy output.

Scripts, raw records, and the same table for all 52 graphs of LiteRT-Models: https://github.com/john-rocky/windows-zoo

Files

file	bytes	sha256
rtdetr_graphA_fp16.onnx	66,894,731	8051234addcc7bc922deab6e312ece2827b57769210c9a2faf08512f4b19926e
openvino/rtdetr_graphA_fp16.xml	1,155,004	2be2d4a287cd40220759c2d2e31f97d39fcf2032c5e7a0eeeec31a39c947a0bf
openvino/rtdetr_graphA_fp16.bin	66,193,692	c9193d60904cfd6718a9458008f7df3e83a6689fb4a044aeeb4cf1738e4cee01
npu/rtdetr_graphA_fp16.blob	40,214,664	39389b96c26d0a23231ce8af28fba81cab2cbc9d8fac2e6c825e768a2d64f40b
rtdetr_graphB_fp16.onnx	15,473,995	8b640ef467b76eea1f9d2c9081b5eebc2de483f38d10d72412e2870541fc469e
openvino/rtdetr_graphB_fp16.xml	1,380,382	b6cab68fac3a123f4c07faf55a717f35ae784706ed8251acf6d98febc1d3e5d3
openvino/rtdetr_graphB_fp16.bin	14,326,260	358ac2d60cfa71d1f874aa35f73d5d2f748e3926243049c11ab12167b8e0df1d
npu/rtdetr_graphB_fp16.blob	76,701,846	245080a4431ed44285c2b9d85c5a2e0b88afb13f275ae2e6cfb89208b0122a69

The NPU blobs were compiled with OpenVINO 2026.4.1, Intel NPU driver 32.0.100.4841 (OpenVINO reports NPU_DRIVER_VERSION 1004841, NPU_COMPILER_VERSION 524292). If import_model refuses a blob on another driver or OpenVINO version, compile the IR instead (the commented line in the snippet).

How it was measured

MSI Stealth 16 AI+ (Core Ultra 9 386H, RTX 5080 Laptop GPU at a 95 W power limit, Windows 11 build 26200.9711, on AC power), 2026-11-07. On the laptop, each row is one process: load, one run, 5 warmup runs, then the median of 30 timed runs (2 warmup + 10 timed runs when the first run takes more than 3 s), numpy input to numpy output, every runtime at its default settings, one process at a time, each started after the CPU, the RTX and the Intel GPU had gone idle (the thresholds are in METHOD.md of the GitHub repo). Output check: each output against the CPU output in fp32 on the same input (ONNX Runtime's CPU EP); PASS = cosine ≥ 0.99, no NaN or Inf, and top-1 agreement ≥ 0.99 for class scores. No Snapdragon rows: Qualcomm AI Hub measured the files before the bias fix (its numbers are in the GitHub table).

rtdetr_graphA_fp16

processor	runtime	file	ms	load s	output check
Intel AI Boost NPU	OpenVINO 2026.4.1	npu/ blob (import_model)	12.07	0.03	PASS (cos ≥ 0.9999)
Intel AI Boost NPU	ONNX Runtime 1.31.1 + OpenVINO EP 1.8.1	.onnx, compiled when the session loads	9.43	10.84	PASS (cos ≥ 0.9999)
Intel Graphics (iGPU)	ONNX Runtime 1.31.1 + OpenVINO EP 1.8.1	.onnx	9.81	1.27	PASS (cos ≥ 0.9999)
RTX 5080 Laptop (95 W limit)	ONNX Runtime 1.31.1, CUDA EP	fp16 copy of the .onnx (not in this repo)	6.54	1.19	PASS (cos ≥ 0.9999)
RTX 5080 Laptop (95 W limit)	ONNX Runtime 1.31.1, CUDA EP	.onnx	7.23	1.21	PASS (cos ≥ 0.9999)
Core Ultra 9 386H CPU	ONNX Runtime 1.31.1, CPU EP	.onnx	58.65	0.33	reference

The ONNX file against the source .tflite (LiteRT interpreter on the CPU, the same input): PASS, cosine ≥ 0.9999 for every output.

rtdetr_graphB_fp16

processor	runtime	file	ms	load s	output check
Intel AI Boost NPU	OpenVINO 2026.4.1	npu/ blob (import_model)	86.90	0.05	PASS (cos ≥ 0.9999)
Intel AI Boost NPU	ONNX Runtime 1.31.1 + OpenVINO EP 1.8.1	.onnx, compiled when the session loads	87.78	65.42	PASS (cos ≥ 0.9999)
Intel Graphics (iGPU)	ONNX Runtime 1.31.1 + OpenVINO EP 1.8.1	.onnx	35.46	1.25	FAIL: output_1_output: argmax agreement 265/300
RTX 5080 Laptop (95 W limit)	ONNX Runtime 1.31.1, CUDA EP	fp16 copy of the .onnx (not in this repo)	8.23	1.19	PASS (cos ≥ 0.9999)
RTX 5080 Laptop (95 W limit)	ONNX Runtime 1.31.1, CUDA EP	.onnx	9.68	0.86	PASS (cos ≥ 0.9999)
Core Ultra 9 386H CPU	ONNX Runtime 1.31.1, CPU EP	.onnx	76.36	0.15	reference

The ONNX file against the source .tflite (LiteRT interpreter on the CPU, the same input): PASS, cosine ≥ 0.9999 for every output.

Run it on the Intel NPU

Download this repo, then run the snippets from its folder:

pip install huggingface_hub
hf download mlboydaisuke/RT-DETRv2-S-OpenVINO --local-dir RT-DETRv2-S-OpenVINO
cd RT-DETRv2-S-OpenVINO

OpenVINO (pip install openvino==2026.4.1):

import numpy as np
import openvino as ov

core = ov.Core()
if "NPU" not in core.available_devices:
    raise SystemExit(f"no Intel NPU visible to OpenVINO: {core.available_devices}")
with open("npu/rtdetr_graphA_fp16.blob", "rb") as f:
    compiled = core.import_model(f.read(), "NPU")  # precompiled: no compile at load
# or compile the IR on this machine: compiled = core.compile_model("openvino/rtdetr_graphA_fp16.xml", "NPU")
# replace the zeros with your preprocessed input (float32, the shapes printed by compiled.inputs)
feeds = {i.get_any_name(): np.zeros(list(i.shape), np.float32) for i in compiled.inputs}
outputs = compiled(feeds)
print({o.get_any_name(): outputs[o].shape for o in compiled.outputs})

ONNX Runtime with the OpenVINO EP (pip install onnxruntime==1.31.1 onnxruntime-ep-openvino==1.8.1):

import numpy as np
import onnxruntime as ort
import onnxruntime_ep_openvino as ov_ep

ort.register_execution_provider_library(ov_ep.get_ep_name(), ov_ep.get_library_path())
npu = [d for d in ort.get_ep_devices()
       if d.ep_name == ov_ep.get_ep_name() and d.ep_metadata.get("ov_device") == "NPU"]
if not npu:
    raise SystemExit("no Intel NPU visible to the OpenVINO EP")
so = ort.SessionOptions()
so.add_provider_for_devices(npu, {})
sess = ort.InferenceSession("rtdetr_graphA_fp16.onnx", sess_options=so)  # compiles for the NPU here
# replace the zeros with your preprocessed input (float32)
feeds = {i.name: np.zeros(i.shape, np.float32) for i in sess.get_inputs()}
outputs = sess.run(None, feeds)
print([o.shape for o in outputs])

The same lines run rtdetr_graphB_fp16 with its own file names.

Licence

Apache-2.0, the licence of litert-community/RT-DETRv2-S-LiteRT. The upstream model: PekingU/rtdetr_v2_r18vd.