Intel/object-classification

Object Classification

Property	Value
Category	Object Classification (Traffic Categorization: People / Vehicles)
Base Model	YOLO26 (Ultralytics, default) or DETR (Facebook, alternative)
Source Framework	PyTorch (Ultralytics, Hugging Face Transformers)
Supported Precisions	FP32, FP16, INT8 (mixed-precision)
Inference Engine	OpenVINO
Hardware	CPU, GPU, NPU
Detected Class(es)	person and vehicle classes grouped into People and Vehicles

---

Overview

Object Classification is a Metro Analytics use case that detects objects with YOLO26 and then categorizes each detection into higher-level city-operations groups. It is built on the state-of-the-art YOLO26 real-time detector, quantized to INT8 for efficient inference on Intel hardware. Where the general object-detection use case reports every one of the 80 COCO classes individually, this use case rolls the traffic-relevant classes up into two semantic categories, People and Vehicles, so operators get an at-a-glance picture of a scene.

The traffic categories are:

• People -- the COCO person class.
• Vehicles -- the COCO bicycle, car, motorcycle, bus, train, and truck classes.

Objects outside these categories are ignored to keep the output focused on traffic situational awareness.

Typical Metro deployments include:

• Situational Awareness -- summarize each camera feed as live People and Vehicles counts.
• Automated City Operations -- feed category counts into signal timing, congestion, and dispatch logic.
• Intersection and Roundabout Monitoring -- track the mix of pedestrians and vehicles at busy junctions.
• Trend Analytics -- aggregate category counts over time to understand traffic patterns.

Available variants: yolo26n, yolo26s, yolo26m, yolo26l. Smaller variants (yolo26n, yolo26s) are recommended for high-FPS edge deployment; larger variants improve recall for small objects.

Model Selection

DETR (DEtection TRansformer) is available as an alternative to YOLO26. DETR is a transformer-based, NMS-free detector trained on COCO and released under the Apache-2.0 license. The export script embeds DETR normalization, box decoding, and the class remap into the OpenVINO IR, so the DETR model produces the same output layout and class ids as YOLO26. The same OpenVINO and DLStreamer samples run either model; select one with the MODEL constant at the top of each sample.

Model	Variants	Input Size	Output	License
YOLO26 (default)	yolo26n, yolo26s, yolo26m, yolo26l	640 x 640	[1, 300, 6]	AGPL-3.0
DETR	detr-resnet-50	800 x 800	[1, 100, 6]	Apache-2.0

Choose YOLO26 for maximum throughput on edge devices. Choose DETR when a permissively licensed transformer detector is preferred; it is heavier and runs at a lower frame rate.

---

Prerequisites

Target stack: Intel DLStreamer 2026.3.0 RC1 with the matching OpenVINO GenAI bundle 2026.5.0.0.dev20261030 (OpenVINO Runtime 2026.5.0.dev20261030). Use the native RC1 setup before running these samples; this is a development stack, not a stable release. See the device validation matrix for measured CPU, GPU, and NPU results.

• Python 3.11+
• Install OpenVINO (latest version)
• Install Intel DLStreamer (latest version)

Create and activate a Python virtual environment before running the scripts:

python3 -m venv .venv --system-site-packages
source .venv/bin/activate

Note: The --system-site-packages flag is required so the virtual environment can access the system-installed OpenVINO and DLStreamer Python packages.

---

Getting Started

Download and Quantize Model

Run the provided script to download, export to OpenVINO IR, and optionally quantize:

chmod +x export_and_quantize.sh
./export_and_quantize.sh

This exports the default yolo26n model in FP16 precision.

Optional: Select a Different Model, Variant, or Precision

./export_and_quantize.sh yolo26n FP32          # full-precision
./export_and_quantize.sh yolo26n INT8          # quantized
./export_and_quantize.sh yolo26s               # larger variant, default FP16
./export_and_quantize.sh detr-resnet-50        # DETR alternative, default FP16
./export_and_quantize.sh detr-resnet-50 INT8   # DETR quantized

The first argument selects the model: any YOLO26 variant (yolo26n, yolo26s, yolo26m, yolo26l) or DETR (detr-resnet-50). The second argument selects the precision (FP32, FP16, INT8); the default is FP16.

The script performs the following steps:

1. Installs dependencies (openvino, ultralytics for YOLO26 or transformers, torch for DETR; adds nncf for INT8).
2. Downloads a sample traffic video (test_video.mp4) of an urban roundabout at low resolution (640x360).
3. Downloads the PyTorch weights and exports to OpenVINO IR.
4. (INT8 only) Quantizes the model using NNCF post-training quantization.

Output files (replace yolo26n with the selected model, for example detr-resnet-50):

• yolo26n_openvino_model/ -- FP32 or FP16 OpenVINO IR model directory.
• yolo26n_objcls_int8.xml / yolo26n_objcls_int8.bin -- INT8 quantized model (only when INT8 is selected).

Precision / Device Compatibility

Precision	CPU	GPU	NPU
FP32	Yes	Yes	No
FP16	Yes	Yes	Yes
INT8	Yes	Yes	Yes

Note: The INT8 calibration uses a frame from the bundled sample video. For production accuracy, replace it with a representative set of frames from the target deployment site.

OpenVINO Sample

The sample below runs YOLO26 inference on the sample traffic video, maps each detection into the People or Vehicles category, draws boxes colored per category, overlays live category counts, and writes the annotated result to output_openvino.mp4. YOLO26 is end-to-end (NMS-free), so no manual non-maximum suppression is needed. Change the device string to run on CPU, GPU, or NPU.

The full sample is in openvino_sample.py; run it with python3 openvino_sample.py or copy the code below.

# SPDX-License-Identifier: MIT
# Copyright (C) Intel Corporation
#
# Object Classification: OpenVINO Runtime sample.
# See README.md for prerequisites and inputs. Run: python3 openvino_sample.py

import cv2
import numpy as np
import openvino as ov

CONF_THRESHOLD = 0.4
# Select the model: "yolo26n" (default) or "detr-resnet-50".
MODEL = "yolo26n"
INPUT_SIZE = 800 if MODEL.startswith("detr") else 640

# Map the traffic-relevant COCO class ids into higher-level city categories.
# People and Vehicles are the two categories tracked for situational awareness.
CATEGORY_BY_CLASS_ID = {
    0: "People",      # person
    1: "Vehicles",    # bicycle
    2: "Vehicles",    # car
    3: "Vehicles",    # motorcycle
    5: "Vehicles",    # bus
    6: "Vehicles",    # train
    7: "Vehicles",    # truck
}
# BGR overlay colors for each category.
CATEGORY_COLORS = {
    "People": (0, 200, 0),
    "Vehicles": (255, 128, 0),
}

core = ov.Core()
model = core.read_model(f"{MODEL}_openvino_model/{MODEL}.xml")

# YOLO26 embeds the 80 COCO class names in rt_info. Ultralytics separates
# multi-word names with underscores (e.g. "traffic_light"), so restore spaces.
COCO_NAMES = [
    name.replace("_", " ")
    for name in model.get_rt_info()["model_info"]["labels"].value.split()
]

# Change device to "GPU" or "NPU" to run on integrated GPU or NPU.
compiled = core.compile_model(model, "CPU")
output_port = compiled.output(0)

cap = cv2.VideoCapture("test_video.mp4")
fps = cap.get(cv2.CAP_PROP_FPS) or 30.0
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
writer = cv2.VideoWriter(
    "output_openvino.mp4", cv2.VideoWriter_fourcc(*"mp4v"), fps, (width, height)
)

totals = {"People": 0, "Vehicles": 0}
frame_idx = 0
while True:
    ok, frame = cap.read()
    if not ok:
        break
    frame_idx += 1

    blob = cv2.resize(frame, (INPUT_SIZE, INPUT_SIZE))
    blob = cv2.cvtColor(blob, cv2.COLOR_BGR2RGB).astype(np.float32) / 255.0
    blob = blob.transpose(2, 0, 1)[np.newaxis, ...]  # NCHW

    # End-to-end output: [1, N, 6] = [x1, y1, x2, y2, confidence, class_id].
    # (N = 300 for YOLO26, 100 for DETR)
    output = compiled([blob])[output_port][0]

    sx, sy = width / INPUT_SIZE, height / INPUT_SIZE
    counts = {"People": 0, "Vehicles": 0}
    for x1, y1, x2, y2, conf, class_id in output:
        if conf < CONF_THRESHOLD:
            continue
        category = CATEGORY_BY_CLASS_ID.get(int(class_id))
        if category is None:
            continue  # not a traffic-relevant object
        counts[category] += 1
        totals[category] += 1
        color = CATEGORY_COLORS[category]
        px1, py1 = int(x1 * sx), int(y1 * sy)
        px2, py2 = int(x2 * sx), int(y2 * sy)
        label = f"{category}: {COCO_NAMES[int(class_id)]} {conf:.2f}"
        cv2.rectangle(frame, (px1, py1), (px2, py2), color, 2)
        cv2.putText(frame, label, (px1, py1 - 5),
                    cv2.FONT_HERSHEY_SIMPLEX, 2.0, color, 2)

    # Overlay the per-category counts for this frame.
    banner = f"People: {counts['People']}  Vehicles: {counts['Vehicles']}"
    cv2.rectangle(frame, (0, 0), (width, 60), (0, 0, 0), -1)
    cv2.putText(frame, banner, (15, 45),
                cv2.FONT_HERSHEY_SIMPLEX, 2.0, (255, 255, 255), 2)

    if frame_idx % 30 == 0:
        print(f"frame {frame_idx}: {banner}", flush=True)

    writer.write(frame)

cap.release()
writer.release()
print(f"Summary: People={totals['People']} Vehicles={totals['Vehicles']}")
print("Saved: output_openvino.mp4")

Device targets:

• "CPU" -- default, works on all Intel platforms.
• "GPU" -- Intel integrated or discrete GPU.
• "NPU" -- Intel NPU (validate with benchmark_app -d NPU).

Try It on a Sample Video

The export_and_quantize.sh script downloads test_video.mp4 automatically. Re-run the OpenVINO sample above. The script reads test_video.mp4, prints the running People and Vehicles counts to the console, and writes the annotated video to output_openvino.mp4.

Expected console output (representative):

frame 30: People: 4  Vehicles: 6
frame 60: People: 3  Vehicles: 7
frame 90: People: 5  Vehicles: 5
Summary: People=372 Vehicles=548
Saved: output_openvino.mp4

Expected Output

DLStreamer Sample

The pipeline below runs the FP16 YOLO26 detector on the sample video via gvadetect, overlays bounding boxes with gvawatermark for the traffic-relevant classes only (non-traffic detections such as handbag are filtered out via show-roi), saves the annotated result to output_dlstreamer.mp4, and prints the People and Vehicles category counts per frame from the detection metadata.

Notes on running this sample:

• Use the FP16 IR (yolo26n_openvino_model/yolo26n.xml or detr-resnet-50_openvino_model/detr-resnet-50.xml). Class names are read automatically from the model's embedded metadata.yaml by DLStreamer 2026.0+ -- no external labels-file is required.

• Export PYTHONPATH so the DLStreamer Python module is importable:

  source /opt/intel/openvino_2026/setupvars.sh
  source /opt/intel/dlstreamer/scripts/setup_dls_env.sh
  export PYTHONPATH=/opt/intel/dlstreamer/python:\
  /opt/intel/dlstreamer/gstreamer/lib/python3/dist-packages:${PYTHONPATH:-}
  

The full sample is in dlstreamer_sample.py; run it with python3 dlstreamer_sample.py or copy the code below.

# SPDX-License-Identifier: MIT
# Copyright (C) Intel Corporation
#
# Object Classification: DLStreamer pipeline sample.
# See README.md for prerequisites and inputs. Run: python3 dlstreamer_sample.py

import gi

gi.require_version("Gst", "1.0")
gi.require_version("GstAnalytics", "1.0")
from gi.repository import Gst, GLib, GstAnalytics

Gst.init([])

INPUT_VIDEO = "test_video.mp4"

# Traffic-relevant COCO labels grouped into higher-level city categories.
CATEGORY_BY_LABEL = {
    "person": "People",
    "bicycle": "Vehicles",
    "car": "Vehicles",
    "motorcycle": "Vehicles",
    "bus": "Vehicles",
    "train": "Vehicles",
    "truck": "Vehicles",
}

# Select the model: "yolo26n" (default) or "detr-resnet-50".
MODEL = "yolo26n"

# For CPU: change device=GPU to device=CPU.
# For NPU: change device=GPU to device=NPU (batch-size=1, nireq=4 recommended).
# gvawatermark displ-cfg:
#   show-roi=... draws only the traffic-relevant classes (person + vehicles),
#     so non-traffic detections such as handbag/backpack are not boxed.
#   font-scale=1.5 enlarges the label text for better visualization.
pipeline_str = (
    f"filesrc location={INPUT_VIDEO} ! decodebin3 ! "
    "videoconvert ! "
    f"gvadetect model={MODEL}_openvino_model/{MODEL}.xml "
    "device=GPU "
    "threshold=0.4 ! queue ! "
    "gvawatermark "
    "displ-cfg=show-roi=person:bicycle:car:motorcycle:bus:train:truck,font-scale=2.5 ! "
    "videoconvert ! video/x-raw,format=I420 ! "
    "openh264enc ! h264parse ! "
    "mp4mux ! filesink name=sink location=output_dlstreamer.mp4"
)
pipeline = Gst.parse_launch(pipeline_str)

totals = {"People": 0, "Vehicles": 0}


def on_buffer(pad, info):
    buf = info.get_buffer()
    rmeta = GstAnalytics.buffer_get_analytics_relation_meta(buf)
    if rmeta is None:
        return Gst.PadProbeReturn.OK
    counts = {"People": 0, "Vehicles": 0}
    idx = 1
    while True:
        ok, od = rmeta.get_od_mtd(idx)
        if not ok:
            break
        label = GLib.quark_to_string(od.get_obj_type())
        category = CATEGORY_BY_LABEL.get(label)
        if category is not None:
            counts[category] += 1
            totals[category] += 1
        idx += 1
    if counts["People"] or counts["Vehicles"]:
        print(f"frame: People={counts['People']} Vehicles={counts['Vehicles']}",
              flush=True)
    return Gst.PadProbeReturn.OK


sink = pipeline.get_by_name("sink")
sink_pad = sink.get_static_pad("sink")
sink_pad.add_probe(Gst.PadProbeType.BUFFER, on_buffer)

pipeline.set_state(Gst.State.PLAYING)
bus = pipeline.get_bus()
bus.timed_pop_filtered(
    Gst.CLOCK_TIME_NONE,
    Gst.MessageType.EOS | Gst.MessageType.ERROR,
)
pipeline.set_state(Gst.State.NULL)
print(f"Summary: People={totals['People']} Vehicles={totals['Vehicles']}")

Expected Output

Device targets:

• device=GPU -- default in the sample code.
• device=CPU -- change device=GPU to device=CPU.
• device=NPU -- change device=GPU to device=NPU; use batch-size=1 and nireq=4 for best NPU utilization.

---

License

Licensed under the MIT License. See LICENSE for details.

References

• YOLO26 Documentation
• DETR Model Card (facebook/detr-resnet-50)
• End-to-End Object Detection with Transformers (DETR paper)
• OpenVINO YOLO26 Notebook
• Sample video: Urban roundabout with cars and pedestrian (Pexels)
• OpenVINO Documentation
• NNCF Post-Training Quantization
• Intel DLStreamer