Port Fidelity: Parity Check

Does LibreYOLO reproduce each detection model's original reported accuracy? Below is an independent re-evaluation on COCO val2017, measured side-by-side with each model's claimed number (sourced from its paper, repo README, model zoo, or HF card).

Parity checked: 05/06/2026Hardware: NVIDIA RTX 5070 TiDataset: COCO val2017 (5000 images)Variants: 61

LibreYOLO PyTorch, fp32, per-image inference at each model's native input size. NMS IoU per family (YOLOX 0.65, YOLOv9 0.7, others 0.6; DETR families are NMS-free). Reference = each model's original paper / repo README / model zoo / HF card claimed COCO box mAP@50-95 (source linked per value).

● |Δ| ≤ 0.3: faithful● ≤ 1.0: minor drift● > 1.0: investigateΔ = measured − claimed (mAP@50-95 points)

YOLOX

Variant	LibreYOLO	Claimed	Δ	Dataset	Source
yolox-nano	25.8	25.8	-0.04	val2017	Megvii-BaseDetection/YOLOX
yolox-tiny	32.7	32.8	-0.14	val2017	Megvii-BaseDetection/YOLOX
yolox-s	40.5	40.5	-0.05	val2017	Megvii-BaseDetection/YOLOX
yolox-m	46.9	46.9	+0.02	val2017	Megvii-BaseDetection/YOLOX
yolox-l	49.8	49.7	+0.05	val2017	Megvii-BaseDetection/YOLOX
yolox-x	51.1	51.1	+0.02	val2017	Megvii-BaseDetection/YOLOX

YOLOv9

Variant	LibreYOLO	Claimed	Δ	Dataset	Source
yolov9t	38.2	38.3	-0.09	val2017	arXiv:2402.13616
yolov9s	46.6	46.8	-0.22	val2017	arXiv:2402.13616
yolov9m	51.2	51.4	-0.24	val2017	arXiv:2402.13616
yolov9c	52.6	53.0	-0.39	val2017	arXiv:2402.13616

RF-DETR

Variant	LibreYOLO	Claimed	Δ	Dataset	Source
rfdetr-n	48.4	48.4	+0.00	val2017	arXiv:2511.09554
rfdetr-s	53.0	53.0	-0.05	val2017	arXiv:2511.09554
rfdetr-m	54.7	54.7	+0.03	val2017	arXiv:2511.09554
rfdetr-l	56.5	56.5	+0.05	val2017	arXiv:2511.09554

RT-DETR

Variant	LibreYOLO	Claimed	Δ	Dataset	Source
rtdetr-r18	46.4	46.5	-0.09	val2017	lyuwenyu/RT-DETR
rtdetr-r34	48.9	48.9	-0.04	val2017	lyuwenyu/RT-DETR
rtdetr-r50m	51.3	51.3	-0.01	val2017	lyuwenyu/RT-DETR
rtdetr-l	52.9	53.0	-0.15	val2017	lyuwenyu/RT-DETR
rtdetr-r50	53.1	53.1	-0.01	val2017	lyuwenyu/RT-DETR
rtdetr-r101	54.4	54.3	+0.05	val2017	lyuwenyu/RT-DETR
rtdetr-x	54.7	54.8	-0.10	val2017	lyuwenyu/RT-DETR

RT-DETRv2

Variant	LibreYOLO	Claimed	Δ	Dataset	Source
rtdetrv2-r18	48.1	48.1	-0.02	val2017	lyuwenyu/RT-DETR
rtdetrv2-r34	49.9	49.9	+0.02	val2017	lyuwenyu/RT-DETR
rtdetrv2-r50m	51.9	51.9	+0.01	val2017	lyuwenyu/RT-DETR
rtdetrv2-r50	53.4	53.4	-0.01	val2017	lyuwenyu/RT-DETR
rtdetrv2-r101	54.4	54.3	+0.05	val2017	lyuwenyu/RT-DETR

RT-DETRv4

Variant	LibreYOLO	Claimed	Δ	Dataset	Source
rtdetrv4-s	49.8	49.7	+0.12	val2017	arXiv:2510.25257
rtdetrv4-m	53.6	53.5	+0.14	val2017	arXiv:2510.25257
rtdetrv4-l	55.4	55.4	+0.02	val2017	arXiv:2510.25257
rtdetrv4-x	57.0	57.0	+0.00	val2017	arXiv:2510.25257

DEIM

Variant	LibreYOLO	Claimed	Δ	Dataset	Source
deim-n	43.0	43.0	+0.05	val2017	Intellindust-AI-Lab/DEIM
deim-s	49.0	49.0	-0.04	val2017	Intellindust-AI-Lab/DEIM
deim-m	52.7	52.7	+0.00	val2017	Intellindust-AI-Lab/DEIM
deim-l	54.7	54.7	+0.04	val2017	Intellindust-AI-Lab/DEIM
deim-x	56.5	56.5	-0.03	val2017	Intellindust-AI-Lab/DEIM

DEIMv2

Variant	LibreYOLO	Claimed	Δ	Dataset	Source
deimv2-atto	23.8	23.8	-0.04	val2017	arXiv:2509.20787
deimv2-femto	31.0	31.0	-0.01	val2017	arXiv:2509.20787
deimv2-pico	38.5	38.5	-0.04	val2017	arXiv:2509.20787
deimv2-n	43.0	43.0	-0.02	val2017	arXiv:2509.20787
deimv2-s	50.9	50.9	-0.04	val2017	arXiv:2509.20787
deimv2-m	53.0	53.0	-0.03	val2017	arXiv:2509.20787
deimv2-l	56.0	56.0	-0.01	val2017	arXiv:2509.20787
deimv2-x	57.8	57.8	+0.01	val2017	arXiv:2509.20787

D-FINE

Variant	LibreYOLO	Claimed	Δ	Dataset	Source
dfine-n	42.8	42.8	-0.01	val2017	Peterande/D-FINE
dfine-s	50.7	50.7	+0.01	val2017	Peterande/D-FINE
dfine-m	55.1	55.1	-0.01	val2017	Peterande/D-FINE
dfine-l	57.3	57.3	-0.05	val2017	Peterande/D-FINE
dfine-x	59.3	59.3	+0.01	val2017	Peterande/D-FINE

PicoDet

Variant	LibreYOLO	Claimed	Δ	Dataset	Source
picodet-s	27.0	27.1	-0.13	val2017	arXiv:2111.00902
picodet-m	34.4	34.3	+0.07	val2017	arXiv:2111.00902
picodet-l	40.6	40.9	-0.34	val2017	arXiv:2111.00902

EC (EdgeCrafter)

Variant	LibreYOLO	Claimed	Δ	Dataset	Source
ec-s	51.7	51.7	+0.01	val2017	arXiv:2603.18739
ec-m	54.2	54.3	-0.08	val2017	arXiv:2603.18739
ec-l	57.0	57.0	-0.01	val2017	arXiv:2603.18739
ec-x	57.9	57.9	-0.02	val2017	arXiv:2603.18739

ONNX Export Fidelity

Does a LibreYOLO ONNX export preserve the same validation behavior as the PyTorch model? This run checks detection, segmentation, and pose exports that claim complete or experimental ONNX support.

Parity checked: 21/06/2026Hardware: NVIDIA GeForce RTX 5070 TiDataset: COCO val2017 mini500Cases: 102

Each case reuses or exports ONNX, validates PyTorch and ONNX on the same 500-image COCO val2017 mini500 dataset, compares task metrics with abs tolerance 0.001 and rel tolerance 0.002, and records the worst metric delta.

Runtime: PyTorch CUDA vs ONNX Runtime CUDAExecutionProvider. Tolerances: abs 0.001, rel 0.002.

Passed

Failed

Unavailable

Total

102

By task

Task	Passed	Failed	Unavailable
detect	68	5	1
segment	5	5	4
pose	6	8	0

By support claim

Claim	Passed	Failed	Unavailable
complete	8	12	4
experimental	71	6	1

damoyolo

6 passed, 0 failed, 1 unavailable

7 cases

damoyolo-l

detectexperimental

unavailable

DAMO-YOLO-L pretrained weights are not available — Alibaba's Aliyun bucket was deleted before any mirror picked them up (see github.com/tinyvision/DAMO-YOLO/issues/144). Use size t, s, m, ns, nm, or nl instead, or build size='l' from scratch for training (LibreDAMOYOLO(size='l').train(allow_experimental=True)).

damoyolo-m

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5497	0.5497	0.00002978	0.001099
mAP50	0.7206	0.7207	0.00009273	0.001441
precision	0.5497	0.5497	0.00002978	0.001099
recallworst	0.7152	0.7160	0.0007299	0.001430

damoyolo-nl

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.4439	0.4442	0.0003193	0.001000
mAP50	0.6205	0.6206	0.0001399	0.001241
precision	0.4439	0.4442	0.0003193	0.001000
recall	0.6104	0.6106	0.0002050	0.001221

damoyolo-nm

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.4144	0.4145	0.0001122	0.001000
mAP50	0.5890	0.5890	0.000001356	0.001178
precision	0.4144	0.4145	0.0001122	0.001000
recallworst	0.5966	0.5965	0.0001384	0.001193

damoyolo-ns

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.3621	0.3621	0.00002840	0.001000
mAP50worst	0.5172	0.5176	0.0004779	0.001034
precision	0.3621	0.3621	0.00002840	0.001000
recall	0.5383	0.5383	0.00002891	0.001077

damoyolo-s

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5035	0.5032	0.0002964	0.001007
mAP50worst	0.6742	0.6738	0.0003677	0.001348
precision	0.5035	0.5032	0.0002964	0.001007
recall	0.6774	0.6775	0.0001276	0.001355

damoyolo-t

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.4662	0.4662	0.00009577	0.001000
mAP50	0.6304	0.6304	0.00002386	0.001261
precision	0.4662	0.4662	0.00009577	0.001000
recallworst	0.6601	0.6603	0.0001941	0.001320

deim

5 passed, 0 failed, 0 unavailable

5 cases

deim-l

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5778	0.5780	0.0001904	0.001156
mAP50	0.7555	0.7556	0.0001472	0.001511
precision	0.5778	0.5780	0.0001904	0.001156
recallworst	0.7602	0.7604	0.0001989	0.001520

deim-m

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5535	0.5538	0.0003190	0.001107
mAP50	0.7196	0.7197	0.00009022	0.001439
precision	0.5535	0.5538	0.0003190	0.001107
recallworst	0.7426	0.7430	0.0004450	0.001485

deim-n

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.4677	0.4681	0.0003819	0.001000
mAP50	0.6378	0.6381	0.0003255	0.001276
precision	0.4677	0.4681	0.0003819	0.001000
recall	0.6631	0.6630	0.0001036	0.001326

deim-s

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.5192	0.5183	0.0009080	0.001038
mAP50	0.6824	0.6816	0.0007318	0.001365
precision	0.5192	0.5183	0.0009080	0.001038
recall	0.7363	0.7355	0.0007980	0.001473

deim-x

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5956	0.5957	0.0001023	0.001191
mAP50worst	0.7695	0.7697	0.0001988	0.001539
precision	0.5956	0.5957	0.0001023	0.001191
recall	0.7703	0.7702	0.00007581	0.001541

deimv2

7 passed, 1 failed, 0 unavailable

8 cases

deimv2-atto

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.2747	0.2752	0.0004963	0.001000
mAP50	0.4031	0.4035	0.0003737	0.001000
precision	0.2747	0.2752	0.0004963	0.001000
recall	0.4687	0.4684	0.0002720	0.001000

deimv2-femto

detectexperimentalworst: mAP50

failed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.3448	0.3454	0.0006443	0.001000
mAP50worst	0.4885	0.4896	0.001053	0.001000
precision	0.3448	0.3454	0.0006443	0.001000
recall	0.5508	0.5518	0.0009602	0.001102

deimv2-l

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5857	0.5858	0.0001027	0.001171
mAP50	0.7561	0.7561	0.000004239	0.001512
precision	0.5857	0.5858	0.0001027	0.001171
recallworst	0.7558	0.7560	0.0001907	0.001512

deimv2-m

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5600	0.5601	0.0001252	0.001120
mAP50worst	0.7292	0.7293	0.0001439	0.001458
precision	0.5600	0.5601	0.0001252	0.001120
recall	0.7368	0.7367	0.00008028	0.001474

deimv2-n

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.4669	0.4672	0.0002626	0.001000
mAP50worst	0.6394	0.6399	0.0004782	0.001279
precision	0.4669	0.4672	0.0002626	0.001000
recall	0.6657	0.6654	0.0003123	0.001331

deimv2-pico

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.4228	0.4228	0.00001438	0.001000
mAP50	0.5931	0.5928	0.0002117	0.001186
precision	0.4228	0.4228	0.00001438	0.001000
recallworst	0.6312	0.6316	0.0004122	0.001262

deimv2-s

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5300	0.5300	0.00003841	0.001060
mAP50worst	0.6977	0.6974	0.0002619	0.001395
precision	0.5300	0.5300	0.00003841	0.001060
recall	0.7239	0.7239	0.000009582	0.001448

deimv2-x

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.6134	0.6134	0.00003692	0.001227
mAP50	0.7896	0.7896	0.00003127	0.001579
precision	0.6134	0.6134	0.00003692	0.001227
recallworst	0.7708	0.7713	0.0004655	0.001542

dfine

4 passed, 1 failed, 0 unavailable

5 cases

dfine-l

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5997	0.5996	0.0001303	0.001199
mAP50	0.7705	0.7703	0.0001910	0.001541
precision	0.5997	0.5996	0.0001303	0.001199
recallworst	0.7831	0.7834	0.0003704	0.001566

dfine-m

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5768	0.5766	0.0002282	0.001154
mAP50	0.7467	0.7462	0.0004907	0.001493
precision	0.5768	0.5766	0.0002282	0.001154
recallworst	0.7699	0.7690	0.0008517	0.001540

dfine-n

detectexperimentalworst: mAP50

failed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.4564	0.4578	0.001467	0.001000
mAP50worst	0.6262	0.6278	0.001576	0.001252
precision	0.4564	0.4578	0.001467	0.001000
recall	0.6494	0.6496	0.0001787	0.001299

dfine-s

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5339	0.5340	0.00006460	0.001068
mAP50worst	0.6979	0.6986	0.0006388	0.001396
precision	0.5339	0.5340	0.00006460	0.001068
recall	0.7339	0.7341	0.0001774	0.001468

dfine-x

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.6130	0.6129	0.0001455	0.001226
mAP50worst	0.7805	0.7800	0.0005576	0.001561
precision	0.6130	0.6129	0.0001455	0.001226
recall	0.7951	0.7950	0.0001344	0.001590

ec

9 passed, 3 failed, 0 unavailable

12 cases

ec-l

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.6011	0.6014	0.0002274	0.001202
mAP50worst	0.7748	0.7752	0.0003070	0.001550
precision	0.6011	0.6014	0.0002274	0.001202
recall	0.7584	0.7583	0.00009409	0.001517

ec-l-pose

poseexperimentalworst: keypoints_mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
keypoints_mAP50-95	0.0006390	0.0007145	0.00007551	0.001000
keypoints_mAP50worst	0.001458	0.001609	0.0001506	0.001000
keypoints_mAP75	0.0005501	0.0006601	0.0001100	0.001000
keypoints_AR50-95	0.01202	0.01202	0	0.001000
keypoints_AR50	0.03686	0.03686	0	0.001000
keypoints_AR75	0.006410	0.006410	0	0.001000

ec-l-segment

segmentexperimentalworst: mAP50(M)

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95(B)	0.5899	0.5900	0.0001235	0.001180
mAP50(B)	0.7705	0.7703	0.0001999	0.001541
precision(B)	0.5899	0.5900	0.0001235	0.001180
recall(B)	0.7524	0.7528	0.0003799	0.001505
mAP50-95(M)	0.5109	0.5111	0.0002003	0.001022
mAP50(M)worst	0.7503	0.7507	0.0004006	0.001501
precision(M)	0.5109	0.5111	0.0002003	0.001022
recall(M)	0.6449	0.6449	0.000005351	0.001290

ec-m

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5818	0.5817	0.00009323	0.001164
mAP50	0.7558	0.7558	0.000003167	0.001512
precision	0.5818	0.5817	0.00009323	0.001164
recallworst	0.7505	0.7503	0.0001706	0.001501

ec-m-pose

poseexperimentalworst: keypoints_AR50-95

failed

Metric	PyTorch	ONNX	Delta	Tolerance
keypoints_mAP50-95	0.04862	0.04855	0.00007334	0.001000
keypoints_mAP50	0.08102	0.08096	0.00005639	0.001000
keypoints_mAP75	0.05098	0.05121	0.0002334	0.001000
keypoints_AR50-95worst	0.4191	0.4176	0.001442	0.001000
keypoints_AR50	0.6426	0.6426	0	0.001285
keypoints_AR75	0.4487	0.4487	0	0.001000

ec-m-segment

segmentexperimentalworst: mAP50(M)

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95(B)	0.5621	0.5619	0.0001589	0.001124
mAP50(B)	0.7392	0.7390	0.0002512	0.001478
precision(B)	0.5621	0.5619	0.0001589	0.001124
recall(B)	0.7324	0.7324	0.00004544	0.001465
mAP50-95(M)	0.4893	0.4892	0.0001383	0.001000
mAP50(M)worst	0.7154	0.7152	0.0002759	0.001431
precision(M)	0.4893	0.4892	0.0001383	0.001000
recall(M)	0.6262	0.6263	0.0001251	0.001252

ec-s

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5436	0.5436	0.00004197	0.001087
mAP50	0.7151	0.7151	0.00002883	0.001430
precision	0.5436	0.5436	0.00004197	0.001087
recallworst	0.7317	0.7318	0.00009273	0.001463

ec-s-pose

poseexperimentalworst: keypoints_AR75

failed

Metric	PyTorch	ONNX	Delta	Tolerance
keypoints_mAP50-95	0.01422	0.01433	0.0001150	0.001000
keypoints_mAP50	0.02368	0.02385	0.0001690	0.001000
keypoints_mAP75	0.01329	0.01325	0.00004755	0.001000
keypoints_AR50-95	0.05208	0.05176	0.0003205	0.001000
keypoints_AR50	0.09295	0.09295	0	0.001000
keypoints_AR75worst	0.04968	0.04808	0.001603	0.001000

ec-s-segment

segmentexperimentalworst: mAP50-95(M)

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95(B)	0.5324	0.5322	0.0001213	0.001065
mAP50(B)	0.7054	0.7053	0.00003833	0.001411
precision(B)	0.5324	0.5322	0.0001213	0.001065
recall(B)	0.7165	0.7167	0.0001864	0.001433
mAP50-95(M)worst	0.4604	0.4601	0.0002299	0.001000
mAP50(M)	0.6857	0.6855	0.0001518	0.001371
precision(M)	0.4604	0.4601	0.0002299	0.001000
recall(M)	0.6083	0.6084	0.0001101	0.001217

ec-x

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.6106	0.6105	0.00004364	0.001221
mAP50worst	0.7852	0.7850	0.0001336	0.001570
precision	0.6106	0.6105	0.00004364	0.001221
recall	0.7616	0.7615	0.00008017	0.001523

ec-x-pose

poseexperimentalworst: keypoints_AR50

failed

Metric	PyTorch	ONNX	Delta	Tolerance
keypoints_mAP50-95	0.001149	0.0009993	0.0001494	0.001000
keypoints_mAP50	0.003871	0.003622	0.0002495	0.001000
keypoints_mAP75	0.0001792	0.0001739	0.000005319	0.001000
keypoints_AR50-95	0.04103	0.04167	0.0006410	0.001000
keypoints_AR50worst	0.1266	0.1298	0.003205	0.001000
keypoints_AR75	0.02244	0.02244	0	0.001000

ec-x-segment

segmentexperimentalworst: recall(M)

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95(B)	0.5978	0.5978	0.00001093	0.001196
mAP50(B)	0.7792	0.7791	0.00006192	0.001558
precision(B)	0.5978	0.5978	0.00001093	0.001196
recall(B)	0.7528	0.7529	0.00008824	0.001506
mAP50-95(M)	0.5236	0.5236	0.00004962	0.001047
mAP50(M)	0.7563	0.7563	0.00001143	0.001513
precision(M)	0.5236	0.5236	0.00004962	0.001047
recall(M)worst	0.6536	0.6541	0.0005113	0.001307

picodet

3 passed, 0 failed, 0 unavailable

3 cases

picodet-l

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.4418	0.4414	0.0004055	0.001000
mAP50	0.6121	0.6121	0.00007665	0.001224
precision	0.4418	0.4414	0.0004055	0.001000
recall	0.6259	0.6259	0.00004496	0.001252

picodet-m

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.3789	0.3788	0.00009854	0.001000
mAP50worst	0.5359	0.5357	0.0002413	0.001072
precision	0.3789	0.3788	0.00009854	0.001000
recall	0.5575	0.5572	0.0002241	0.001115

picodet-s

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.3039	0.3037	0.0002206	0.001000
mAP50	0.4426	0.4426	0.00004185	0.001000
precision	0.3039	0.3037	0.0002206	0.001000
recall	0.4803	0.4804	0.00008880	0.001000

rfdetr

4 passed, 11 failed, 0 unavailable

15 cases

rfdetr-l

detectcompleteworst: recall

failed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5853	0.5861	0.0007269	0.001171
mAP50	0.7602	0.7610	0.0008061	0.001520
precision	0.5853	0.5861	0.0007269	0.001171
recallworst	0.7596	0.7614	0.001740	0.001519

rfdetr-l-pose

posecompleteworst: keypoints_AR50

failed

Metric	PyTorch	ONNX	Delta	Tolerance
keypoints_mAP50-95	0.000009108	0.000008844	2.645e-7	0.001000
keypoints_mAP50	0.00002273	0.00002300	2.641e-7	0.001000
keypoints_mAP75	0.000006929	0.000006596	3.324e-7	0.001000
keypoints_AR50-95	0.004327	0.003526	0.0008013	0.001000
keypoints_AR50worst	0.008013	0.006410	0.001603	0.001000
keypoints_AR75	0.003205	0.003205	0	0.001000

rfdetr-l-segment

segmentcompleteworst: recall(M)

failed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95(B)	0.5802	0.5812	0.0009621	0.001160
mAP50(B)	0.7569	0.7584	0.001471	0.001514
precision(B)	0.5802	0.5812	0.0009621	0.001160
recall(B)	0.7521	0.7535	0.001359	0.001504
mAP50-95(M)	0.4982	0.4974	0.0007724	0.001000
mAP50(M)	0.7308	0.7310	0.0001799	0.001462
precision(M)	0.4982	0.4974	0.0007724	0.001000
recall(M)worst	0.6390	0.6370	0.002032	0.001278

rfdetr-m

detectcompleteworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.5731	0.5740	0.0008728	0.001146
mAP50	0.7528	0.7528	0.00002819	0.001506
precision	0.5731	0.5740	0.0008728	0.001146
recall	0.7421	0.7427	0.0006034	0.001484

rfdetr-m-pose

posecompleteworst: keypoints_AR50

failed

Metric	PyTorch	ONNX	Delta	Tolerance
keypoints_mAP50-95	0.00002346	0.00002679	0.000003324	0.001000
keypoints_mAP50	0.00004674	0.00005943	0.00001269	0.001000
keypoints_mAP75	0.00002427	0.00002195	0.000002314	0.001000
keypoints_AR50-95	0.006731	0.007051	0.0003205	0.001000
keypoints_AR50worst	0.01122	0.01282	0.001603	0.001000
keypoints_AR75	0.006410	0.006410	0	0.001000

rfdetr-m-segment

segmentcompleteworst: mAP50(M)

failed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95(B)	0.5675	0.5697	0.002196	0.001135
mAP50(B)	0.7449	0.7466	0.001731	0.001490
precision(B)	0.5675	0.5697	0.002196	0.001135
recall(B)	0.7435	0.7455	0.001936	0.001487
mAP50-95(M)	0.4831	0.4844	0.001259	0.001000
mAP50(M)worst	0.7149	0.7171	0.002204	0.001430
precision(M)	0.4831	0.4844	0.001259	0.001000
recall(M)	0.6205	0.6197	0.0007346	0.001241

rfdetr-n

detectcompleteworst: recall

failed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5136	0.5136	0.00001603	0.001027
mAP50	0.6990	0.6981	0.0008661	0.001398
precision	0.5136	0.5136	0.00001603	0.001027
recallworst	0.6729	0.6743	0.001461	0.001346

rfdetr-n-pose

posecompleteworst: keypoints_AR50

failed

Metric	PyTorch	ONNX	Delta	Tolerance
keypoints_mAP50-95	0.00002188	0.00002241	5.327e-7	0.001000
keypoints_mAP50	0.00002978	0.00003042	6.404e-7	0.001000
keypoints_mAP75	0.00001750	0.00001715	3.537e-7	0.001000
keypoints_AR50-95	0.004487	0.005288	0.0008013	0.001000
keypoints_AR50worst	0.004808	0.006410	0.001603	0.001000
keypoints_AR75	0.004808	0.004808	0	0.001000

rfdetr-n-segment

segmentcompleteworst: mAP50(M)

failed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95(B)	0.5264	0.5274	0.0009850	0.001053
mAP50(B)	0.7082	0.7093	0.001036	0.001416
precision(B)	0.5264	0.5274	0.0009850	0.001053
recall(B)	0.6701	0.6700	0.0001306	0.001340
mAP50-95(M)	0.4401	0.4409	0.0007463	0.001000
mAP50(M)worst	0.6656	0.6679	0.002282	0.001331
precision(M)	0.4401	0.4409	0.0007463	0.001000
recall(M)	0.5500	0.5500	0.00005527	0.001100

rfdetr-s

detectcompleteworst: recall

failed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5511	0.5513	0.0002106	0.001102
mAP50	0.7346	0.7343	0.0003705	0.001469
precision	0.5511	0.5513	0.0002106	0.001102
recallworst	0.7213	0.7246	0.003272	0.001443

rfdetr-s-pose

posecompleteworst: keypoints_mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
keypoints_mAP50-95worst	0.00001504	0.00001563	5.865e-7	0.001000
keypoints_mAP50	0.00003075	0.00003084	9.579e-8	0.001000
keypoints_mAP75	0.00002116	0.00002148	3.212e-7	0.001000
keypoints_AR50-95	0.004167	0.004167	8.674e-19	0.001000
keypoints_AR50	0.006410	0.006410	0	0.001000
keypoints_AR75	0.004808	0.004808	0	0.001000

rfdetr-s-segment

segmentcompleteworst: recall(B)

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95(B)	0.5409	0.5410	0.00003780	0.001082
mAP50(B)	0.7175	0.7172	0.0002921	0.001435
precision(B)	0.5409	0.5410	0.00003780	0.001082
recall(B)worst	0.6901	0.6915	0.001378	0.001380
mAP50-95(M)	0.4599	0.4597	0.0002243	0.001000
mAP50(M)	0.6892	0.6902	0.0009527	0.001378
precision(M)	0.4599	0.4597	0.0002243	0.001000
recall(M)	0.5784	0.5783	0.00003337	0.001157

rfdetr-x-pose

posecompleteworst: keypoints_AR50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
keypoints_mAP50-95	0.7164	0.7154	0.0009883	0.001433
keypoints_mAP50	0.9180	0.9179	0.0001212	0.001836
keypoints_mAP75	0.7899	0.7902	0.0002200	0.001580
keypoints_AR50-95	0.7796	0.7788	0.0008013	0.001559
keypoints_AR50worst	0.9519	0.9535	0.001603	0.001904
keypoints_AR75	0.8558	0.8574	0.001603	0.001712

rfdetr-x-segment

segmentcompleteworst: recall(B)

failed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95(B)	0.5927	0.5928	0.0001096	0.001185
mAP50(B)	0.7691	0.7685	0.0006010	0.001538
precision(B)	0.5927	0.5928	0.0001096	0.001185
recall(B)worst	0.7771	0.7795	0.002337	0.001554
mAP50-95(M)	0.5175	0.5169	0.0005888	0.001035
mAP50(M)	0.7452	0.7448	0.0004279	0.001490
precision(M)	0.5175	0.5169	0.0005888	0.001035
recall(M)	0.6723	0.6713	0.001022	0.001345

rfdetr-xx-segment

segmentcompleteworst: mAP50(M)

failed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95(B)	0.6061	0.6103	0.004190	0.001212
mAP50(B)	0.7770	0.7816	0.004602	0.001554
precision(B)	0.6061	0.6103	0.004190	0.001212
recall(B)	0.7887	0.7883	0.0004601	0.001577
mAP50-95(M)	0.5298	0.5328	0.003075	0.001060
mAP50(M)worst	0.7571	0.7624	0.005276	0.001514
precision(M)	0.5298	0.5328	0.003075	0.001060
recall(M)	0.6845	0.6842	0.0002984	0.001369

rtdetr

7 passed, 0 failed, 0 unavailable

7 cases

rtdetr-l

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5578	0.5578	0.000006037	0.001116
mAP50	0.7370	0.7369	0.00004305	0.001474
precision	0.5578	0.5578	0.000006037	0.001116
recallworst	0.7415	0.7417	0.0001786	0.001483

rtdetr-r101

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.5661	0.5664	0.0003012	0.001132
mAP50	0.7374	0.7377	0.0002659	0.001475
precision	0.5661	0.5664	0.0003012	0.001132
recall	0.7543	0.7546	0.0002550	0.001509

rtdetr-r18

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.4986	0.4987	0.0001705	0.001000
mAP50	0.6647	0.6646	0.0001284	0.001329
precision	0.4986	0.4987	0.0001705	0.001000
recallworst	0.7226	0.7229	0.0003782	0.001445

rtdetr-r34

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5214	0.5213	0.00004458	0.001043
mAP50worst	0.6952	0.6950	0.0001562	0.001390
precision	0.5214	0.5213	0.00004458	0.001043
recall	0.7215	0.7214	0.00004261	0.001443

rtdetr-r50

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5587	0.5586	0.00008320	0.001117
mAP50worst	0.7311	0.7308	0.0002363	0.001462
precision	0.5587	0.5586	0.00008320	0.001117
recall	0.7462	0.7461	0.0001315	0.001492

rtdetr-r50m

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.5379	0.5382	0.0002639	0.001076
mAP50	0.7114	0.7116	0.0001900	0.001423
precision	0.5379	0.5382	0.0002639	0.001076
recall	0.7298	0.7299	0.0001041	0.001460

rtdetr-x

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5793	0.5790	0.0002877	0.001159
mAP50worst	0.7595	0.7591	0.0004295	0.001519
precision	0.5793	0.5790	0.0002877	0.001159
recall	0.7617	0.7614	0.0003089	0.001523

rtdetrv2

5 passed, 0 failed, 0 unavailable

5 cases

rtdetrv2-r101

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.5681	0.5685	0.0004048	0.001136
mAP50	0.7398	0.7401	0.0003382	0.001480
precision	0.5681	0.5685	0.0004048	0.001136
recall	0.7548	0.7548	0.00005185	0.001510

rtdetrv2-r18

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5078	0.5079	0.00007052	0.001016
mAP50	0.6744	0.6745	0.00005996	0.001349
precision	0.5078	0.5079	0.00007052	0.001016
recallworst	0.7275	0.7276	0.0001705	0.001455

rtdetrv2-r34

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5323	0.5325	0.0001868	0.001065
mAP50	0.7024	0.7025	0.0001076	0.001405
precision	0.5323	0.5325	0.0001868	0.001065
recallworst	0.7203	0.7205	0.0002096	0.001441

rtdetrv2-r50

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5577	0.5576	0.0001275	0.001115
mAP50worst	0.7286	0.7284	0.0001637	0.001457
precision	0.5577	0.5576	0.0001275	0.001115
recall	0.7416	0.7415	0.0001014	0.001483

rtdetrv2-r50m

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5475	0.5474	0.00008117	0.001095
mAP50	0.7179	0.7178	0.00009237	0.001436
precision	0.5475	0.5474	0.00008117	0.001095
recallworst	0.7411	0.7401	0.0009955	0.001482

rtdetrv4

4 passed, 0 failed, 0 unavailable

4 cases

rtdetrv4-l

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.5779	0.5777	0.0002454	0.001156
mAP50	0.7447	0.7445	0.0001195	0.001489
precision	0.5779	0.5777	0.0002454	0.001156
recall	0.7648	0.7648	0.00007179	0.001530

rtdetrv4-m

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5648	0.5639	0.0008959	0.001130
mAP50worst	0.7298	0.7288	0.0009896	0.001460
precision	0.5648	0.5639	0.0008959	0.001130
recall	0.7489	0.7490	0.0001453	0.001498

rtdetrv4-s

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5276	0.5280	0.0004874	0.001055
mAP50	0.6982	0.6986	0.0003672	0.001396
precision	0.5276	0.5280	0.0004874	0.001055
recallworst	0.7299	0.7292	0.0007045	0.001460

rtdetrv4-x

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5992	0.5991	0.0001272	0.001198
mAP50worst	0.7711	0.7709	0.0001725	0.001542
precision	0.5992	0.5991	0.0001272	0.001198
recall	0.7758	0.7758	0.000007514	0.001552

rtmdet

5 passed, 0 failed, 0 unavailable

5 cases

rtmdet-l

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5402	0.5404	0.0001303	0.001080
mAP50	0.7131	0.7132	0.00007577	0.001426
precision	0.5402	0.5404	0.0001303	0.001080
recallworst	0.7126	0.7129	0.0002374	0.001425

rtmdet-m

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.5275	0.5273	0.0001886	0.001055
mAP50	0.6959	0.6958	0.0001090	0.001392
precision	0.5275	0.5273	0.0001886	0.001055
recall	0.6926	0.6925	0.0001285	0.001385

rtmdet-s

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.4616	0.4626	0.0009853	0.001000
mAP50worst	0.6339	0.6352	0.001207	0.001268
precision	0.4616	0.4626	0.0009853	0.001000
recall	0.6446	0.6444	0.0002630	0.001289

rtmdet-t

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.4375	0.4373	0.0001608	0.001000
mAP50	0.6105	0.6106	0.0001142	0.001221
precision	0.4375	0.4373	0.0001608	0.001000
recallworst	0.6337	0.6333	0.0003192	0.001267

rtmdet-x

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.5772	0.5771	0.0001114	0.001154
mAP50	0.7477	0.7479	0.0001017	0.001495
precision	0.5772	0.5771	0.0001114	0.001154
recall	0.7306	0.7307	0.00006952	0.001461

yolo9

4 passed, 1 failed, 4 unavailable

9 cases

yolo9-c

detectcompleteworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5709	0.5709	0.00009272	0.001142
mAP50worst	0.7382	0.7387	0.0004285	0.001476
precision	0.5709	0.5709	0.00009272	0.001142
recall	0.7197	0.7198	0.0001417	0.001439

yolo9-c-segment

segmentcomplete

unavailable

Failed to download weights from https://huggingface.co/LibreYOLO/LibreYOLO9c-seg/resolve/main/LibreYOLO9c-seg.pt: 401 Client Error: Unauthorized for url: https://huggingface.co/LibreYOLO/LibreYOLO9c-seg/resolve/main/LibreYOLO9c-seg.pt

yolo9-m

detectcompleteworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5611	0.5611	0.00001605	0.001122
mAP50	0.7266	0.7266	0.00002263	0.001453
precision	0.5611	0.5611	0.00001605	0.001122
recallworst	0.7144	0.7140	0.0003481	0.001429

yolo9-m-segment

segmentcomplete

unavailable

Failed to download weights from https://huggingface.co/LibreYOLO/LibreYOLO9m-seg/resolve/main/LibreYOLO9m-seg.pt: 401 Client Error: Unauthorized for url: https://huggingface.co/LibreYOLO/LibreYOLO9m-seg/resolve/main/LibreYOLO9m-seg.pt

yolo9-s

detectcompleteworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.5045	0.5042	0.0002415	0.001009
mAP50	0.6720	0.6718	0.0002025	0.001344
precision	0.5045	0.5042	0.0002415	0.001009
recall	0.6777	0.6775	0.0001547	0.001355

yolo9-s-pose

posecompleteworst: keypoints_mAP75

failed

Metric	PyTorch	ONNX	Delta	Tolerance
keypoints_mAP50-95	0.4422	0.4418	0.0003373	0.001000
keypoints_mAP50	0.7950	0.7950	0.00003233	0.001590
keypoints_mAP75worst	0.4290	0.4265	0.002523	0.001000
keypoints_AR50-95	0.5210	0.5207	0.0003205	0.001042
keypoints_AR50	0.8446	0.8446	0	0.001689
keypoints_AR75	0.5433	0.5417	0.001603	0.001087

yolo9-s-segment

segmentcomplete

unavailable

Failed to download weights from https://huggingface.co/LibreYOLO/LibreYOLO9s-seg/resolve/main/LibreYOLO9s-seg.pt: 401 Client Error: Unauthorized for url: https://huggingface.co/LibreYOLO/LibreYOLO9s-seg/resolve/main/LibreYOLO9s-seg.pt

yolo9-t

detectcompleteworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.4177	0.4180	0.0002666	0.001000
mAP50	0.5663	0.5664	0.0001725	0.001133
precision	0.4177	0.4180	0.0002666	0.001000
recallworst	0.6209	0.6219	0.001036	0.001242

yolo9-t-segment

segmentcomplete

unavailable

Failed to download weights from https://huggingface.co/LibreYOLO/LibreYOLO9t-seg/resolve/main/LibreYOLO9t-seg.pt: 401 Client Error: Unauthorized for url: https://huggingface.co/LibreYOLO/LibreYOLO9t-seg/resolve/main/LibreYOLO9t-seg.pt

yolo9_e2e

4 passed, 0 failed, 0 unavailable

4 cases

yolo9_e2e-c

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.5498	0.5499	0.0001395	0.001100
mAP50	0.7105	0.7104	0.0001081	0.001421
precision	0.5498	0.5499	0.0001395	0.001100
recall	0.7239	0.7240	0.0001367	0.001448

yolo9_e2e-m

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.5300	0.5298	0.0001736	0.001060
mAP50	0.6967	0.6965	0.0001680	0.001393
precision	0.5300	0.5298	0.0001736	0.001060
recall	0.7148	0.7148	0.00004116	0.001430

yolo9_e2e-s

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.4677	0.4676	0.00006111	0.001000
mAP50	0.6265	0.6267	0.0001579	0.001253
precision	0.4677	0.4676	0.00006111	0.001000
recallworst	0.6610	0.6608	0.0001772	0.001322

yolo9_e2e-t

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.3870	0.3869	0.00009888	0.001000
mAP50	0.5224	0.5224	0.00005743	0.001045
precision	0.3870	0.3869	0.00009888	0.001000
recall	0.6221	0.6221	0.00002078	0.001244

yolonas

6 passed, 1 failed, 0 unavailable

7 cases

yolonas-l

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5630	0.5631	0.00008586	0.001126
mAP50	0.7327	0.7327	0.00003790	0.001465
precision	0.5630	0.5631	0.00008586	0.001126
recallworst	0.6914	0.6917	0.0002329	0.001383

yolonas-l-pose

poseexperimentalworst: keypoints_AR50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
keypoints_mAP50-95	0.6822	0.6822	0.00003099	0.001364
keypoints_mAP50	0.9037	0.9036	0.00008399	0.001807
keypoints_mAP75	0.7508	0.7508	0.000006702	0.001502
keypoints_AR50-95worst	0.7287	0.7288	0.0001603	0.001457
keypoints_AR50	0.9327	0.9327	0	0.001865
keypoints_AR75	0.7933	0.7933	0	0.001587

yolonas-m

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5544	0.5543	0.00002196	0.001109
mAP50worst	0.7224	0.7225	0.0001301	0.001445
precision	0.5544	0.5543	0.00002196	0.001109
recall	0.6929	0.6929	0.00003827	0.001386

yolonas-m-pose

poseexperimentalworst: keypoints_mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
keypoints_mAP50-95worst	0.6686	0.6689	0.0003096	0.001337
keypoints_mAP50	0.8842	0.8842	0.00003369	0.001768
keypoints_mAP75	0.7333	0.7333	0.00006449	0.001467
keypoints_AR50-95	0.7223	0.7224	0.0001603	0.001445
keypoints_AR50	0.9151	0.9151	0	0.001830
keypoints_AR75	0.7821	0.7821	0	0.001564

yolonas-n-pose

poseexperimentalworst: keypoints_mAP75

failed

Metric	PyTorch	ONNX	Delta	Tolerance
keypoints_mAP50-95	0.5988	0.5989	0.00005189	0.001198
keypoints_mAP50	0.8406	0.8406	0.000003515	0.001681
keypoints_mAP75worst	0.6555	0.6569	0.001456	0.001311
keypoints_AR50-95	0.6537	0.6535	0.0001603	0.001307
keypoints_AR50	0.8814	0.8814	0	0.001763
keypoints_AR75	0.7115	0.7115	0	0.001423

yolonas-s

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5176	0.5175	0.00008679	0.001035
mAP50	0.6900	0.6900	0.000007492	0.001380
precision	0.5176	0.5175	0.00008679	0.001035
recallworst	0.6578	0.6579	0.0001686	0.001316

yolonas-s-pose

poseexperimentalworst: keypoints_AR50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
keypoints_mAP50-95	0.6377	0.6376	0.00008350	0.001275
keypoints_mAP50	0.8597	0.8598	0.00006025	0.001719
keypoints_mAP75	0.6934	0.6934	0.000007695	0.001387
keypoints_AR50-95worst	0.6981	0.6978	0.0003205	0.001396
keypoints_AR50	0.9022	0.9022	0	0.001804
keypoints_AR75	0.7516	0.7516	0	0.001503

yolox

6 passed, 0 failed, 0 unavailable

6 cases

yolox-l

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5537	0.5539	0.0001534	0.001107
mAP50	0.7299	0.7301	0.0001862	0.001460
precision	0.5537	0.5539	0.0001534	0.001107
recallworst	0.6834	0.6836	0.0001937	0.001367

yolox-m

detectexperimentalworst: mAP50-95

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95worst	0.5169	0.5175	0.0005291	0.001034
mAP50	0.7017	0.7018	0.0001114	0.001403
precision	0.5169	0.5175	0.0005291	0.001034
recall	0.6666	0.6669	0.0002377	0.001333

yolox-n

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.2877	0.2885	0.0007475	0.001000
mAP50	0.4474	0.4481	0.0006610	0.001000
precision	0.2877	0.2885	0.0007475	0.001000
recallworst	0.4385	0.4393	0.0008640	0.001000

yolox-s

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.4426	0.4428	0.0001514	0.001000
mAP50worst	0.6298	0.6295	0.0003450	0.001260
precision	0.4426	0.4428	0.0001514	0.001000
recall	0.6010	0.6009	0.00002670	0.001202

yolox-t

detectexperimentalworst: mAP50

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.3594	0.3594	0.00004663	0.001000
mAP50worst	0.5399	0.5401	0.0001747	0.001080
precision	0.3594	0.3594	0.00004663	0.001000
recall	0.5041	0.5042	0.00007256	0.001008

yolox-x

detectexperimentalworst: recall

passed

Metric	PyTorch	ONNX	Delta	Tolerance
mAP50-95	0.5623	0.5625	0.0002077	0.001125
mAP50	0.7427	0.7428	0.00009598	0.001485
precision	0.5623	0.5625	0.0002077	0.001125
recallworst	0.6905	0.6908	0.0002858	0.001381

Claimed values are each model's own published number on COCO and are not LibreYOLO results; some are reported on test-dev rather than val2017 (see the Dataset column). A small residual is expected from fp16 weight storage and minor pre/post-processing differences. Full per-variant results live in the benchmark tables.