YOLOX-Pylon-L

A YOLOX-L detector extended with one additional class — traffic cones (traffic_cone) — on top of the 80 COCO classes, for 81 classes total. The adaptation is trained so the original COCO capabilities are kept, not traded away: this checkpoint retains 97% of the official YOLOX-L baseline while the added cone class scores higher than 79 of the 80 original classes.

Cones are the public demo class. The same adaptation recipe adds arbitrary custom classes (defects, parts, PPE, and similar) to a proven detector without losing what it already knows.

Part of the YOLOX-Pylon family: S · M · L · X (in training).

Built by Empirisch Tech GmbH (Vienna, Austria) under the Chaperone AI brand — see About Empirisch Tech below.

Results

Evaluated on COCO val2017 plus a held-out traffic-cone split, 81 classes in a single pass, 640×640 input, IoU 0.50:0.95 unless noted.

Metric	Value
mAP 50:95 (81 classes)	48.9
mAP 50:95, original 80 COCO classes only	48.5
AP50 / AP75	66.0 / 52.7
AP small / medium / large	31.1 / 53.2 / 63.5
AR@100	60.8
traffic_cone AP / AR	78.6 / 81.4
Inference (forward + NMS, batch 1, fp16)	3.71 ms

Two things worth noting:

• Retention held at 97%. The official YOLOX-L baseline is 50.1 mAP on COCO. After adding the cone class, this checkpoint keeps 48.5 on the same 80 classes — 1.6 points traded for an entire new class arriving at top-tier accuracy.
• The added class outscores 79 of 80 originals. At 78.6 AP, traffic_cone is second only to bear (79.7) on this checkpoint, ahead of giraffe (76.4), stop sign (75.8), and far ahead of workhorse classes like person (61.6) and car (58.5).

Full per-class AP (81 classes)

class	AP	class	AP	class	AP
person	61.618	bicycle	37.547	car	58.494
motorcycle	51.265	airplane	73.783	bus	74.632
train	71.065	truck	53.815	boat	30.934
traffic light	43.080	fire hydrant	75.738	stop sign	75.843
parking meter	51.456	bench	37.430	bird	41.997
cat	74.731	dog	67.190	horse	67.541
sheep	59.677	cow	63.133	elephant	71.051
bear	79.689	zebra	74.188	giraffe	76.406
backpack	20.183	umbrella	47.879	handbag	20.090
tie	43.556	suitcase	48.253	frisbee	69.678
skis	33.713	snowboard	41.683	sports ball	50.590
kite	51.097	baseball bat	39.119	baseball glove	41.565
skateboard	62.194	surfboard	45.487	tennis racket	59.014
bottle	42.903	wine glass	41.440	cup	48.639
fork	49.134	knife	26.992	spoon	26.324
bowl	46.760	banana	28.472	apple	19.796
sandwich	35.887	orange	31.618	broccoli	24.728
carrot	27.921	hot dog	43.810	pizza	59.750
donut	50.739	cake	43.368	chair	38.982
couch	49.349	potted plant	30.830	bed	49.074
dining table	30.917	toilet	64.619	tv	61.952
laptop	68.424	mouse	64.775	remote	43.296
keyboard	56.172	cell phone	44.216	microwave	69.574
oven	43.304	toaster	28.361	sink	43.081
refrigerator	62.442	book	16.598	clock	52.497
vase	41.479	scissors	34.524	teddy bear	52.537
hair drier	10.184	toothbrush	29.280	traffic_cone	78.623

Comparison with other detectors

Large-tier detectors, published COCO val2017 figures from the official YOLOX and Ultralytics model tables.

Model	COCO mAP 50:95	Params	Custom classes	License
yolox-pylon-l (this model)	48.5 kept + traffic_cone 78.6	54.2M	yours added, COCO kept	Apache-2.0
YOLOX-L (base)	50.1	54.2M	COCO only	Apache-2.0
YOLO11l	53.4	25.3M	COCO only	AGPL-3.0 / commercial
YOLO26l	55.0	24.8M	COCO only	AGPL-3.0 / commercial

How to read this honestly: the newest Ultralytics releases post higher raw COCO scores at half the parameters — several years of architecture progress. This model optimizes for a different job: extending a commercially permissive base with new classes while retaining its original capabilities. The relevant scores are the retention delta (1.6 against its own baseline) and the added-class AP (78.6), not the raw COCO leaderboard. The Apache-2.0 license also means the weights can be deployed commercially without a per-deployment license or an obligation to open-source derivative work, which AGPL-3.0 models require.

Siblings for scale (same recipe, same eval protocol):

Family member	mAP (81 cls)	COCO kept	Cone AP	Inference
yolox-pylon-s	42.0	41.6	74.5	1.7 ms
yolox-pylon-m	47.2	46.8	77.5	2.6 ms
yolox-pylon-l	48.9	48.5	78.6	3.7 ms
yolox-pylon-x	in training	—	—	—

Usage

The checkpoint loads with the official YOLOX codebase. The only change from stock YOLOX-L is num_classes = 81, with traffic_cone as class index 80.

import torch
from yolox.exp import get_exp
from yolox.utils import postprocess

# stock yolox-l exp, patched to 81 classes
exp = get_exp(exp_name="yolox-l")
exp.num_classes = 81

model = exp.get_model()
ckpt = torch.load("yolox-pylon-l.pth", map_location="cpu")
model.load_state_dict(ckpt["model"])
model.eval().cuda()

# img: float32 tensor [1, 3, 640, 640], preprocessed YOLOX-style
with torch.no_grad():
    outputs = model(img)
outputs = postprocess(outputs, num_classes=81, conf_thre=0.25, nms_thre=0.45)

COCO_CLASSES = [...]                        # standard 80-class list
CLASSES = COCO_CLASSES + ["traffic_cone"]   # index 80

Or with the repo's demo tool:

git clone https://github.com/Megvii-BaseDetection/YOLOX && cd YOLOX
python tools/demo.py image \
    -f exps/default/yolox_l.py \
    -c yolox-pylon-l.pth \
    --path your_image.jpg --conf 0.25 --nms 0.45 --tsize 640 --device gpu
# patch exps/default/yolox_l.py with self.num_classes = 81 first

Training

• Base: YOLOX-L (54.2M params), initialized from COCO-pretrained weights
• Data: COCO train2017 plus a labeled traffic-cone dataset, trained jointly so the original 80 classes stay in the mix during adaptation
• Eval: COCO val2017 plus a held-out cone split, single 81-class evaluation pass
• Input: 640×640

Intended use and limitations

Intended for roadside and infrastructure perception where traffic cones matter (work zones, lane closures, autonomous driving research) and as a template for class-extension on YOLOX. The L size is the accuracy pick of the released family — strongest on every bucket, including small objects (31.1 AP) — and fits server-side inference over multiple camera streams, as well as offline analysis and auto-labeling. The model detects boxes for 81 classes; it does not segment, track, or estimate distance. For fixed single-camera deployments, yolox-pylon-m offers most of the accuracy at 2.6 ms; for embedded and edge boards, yolox-pylon-s runs in 1.7 ms. As with any detector, validate on your own cameras before production use.

About Empirisch Tech

YOLOX-Pylon is built by Empirisch Tech GmbH, a Vienna-based AI company, under its Chaperone AI brand. The company runs one recipe across three domains — adapt a proven foundation model to a specific domain, keep what the base already knows, and ship the checkpoint together with the data it was trained on:

• Language — Thinking-LQ-1.0 (84% MedQA, within 4 points of GPT-4o at ~20GB) and Coder-LQ-1.0
• Physics — Chaperone-Flow-1.0 (Poseidon-B extended to new CFD regimes, 1.8% wake error) and Palace-LoRA (electromagnetics solver configs)
• Vision — the YOLOX-Pylon family and a road-scene anomaly segmentation pipeline

The models power the company's production platforms, including NumericalAI (GPU physics simulation) and Simvera (industrial perception trained in simulation, deployed on real cameras). Everything is self-hosted in the company's own Vienna datacenter — no third-party model APIs. Empirisch Tech is a member of the NVIDIA Inception and Microsoft for Startups programs, and its open checkpoints have passed 30,000 downloads on Hugging Face.

Custom builds: the cone class took one adaptation run. For your own classes, cameras, or datasets, reach out via chaperoneai.com/contact.

License

Apache-2.0, matching the YOLOX base.

Citation

@article{yolox2021,
  title={YOLOX: Exceeding YOLO Series in 2021},
  author={Ge, Zheng and Liu, Songtao and Wang, Feng and Li, Zeming and Sun, Jian},
  journal={arXiv preprint arXiv:2107.08430},
  year={2021}
}