yolo26_classify

YOLO26-cls: Optimized for AMD ROCm

YOLO26-cls is a real-time image classification model that predicts a class over the 1000 ImageNet categories in a single forward pass. This repository packages inference for image classification using ONNX Runtime, exported and validated for AMD ROCm so it runs efficiently on AMD GPUs, CPUs, and NPUs.

This is based on the implementation of YOLO26 found here. This repository contains configurations and scripts optimized for AMD® ROCm™ platforms. You can use the yolo26_classify AMD scripts to reproduce results or export with custom configurations. More details on model performance can be found here.

---

Task Overview

Task: Image classification

Dataset: ImageNet-1000 label space (a sample set — the imagenette subset — is staged under datasets/samples/ for visual evaluation)

Output metrics: Throughput (inferences/sec), latency (mean/P95/P99 ms), per-operator profiling breakdown

Model variants: Default is n (nano). Override with MODEL_SIZE=n/s/m/l/x — note the classify download list ships n/s/m/l weights only, so MODEL_SIZE=x may not resolve to a hosted weight.

---

AMD ROCm Optimization

This model export has been adapted and validated for AMD Instinct™ / Radeon™ GPUs running ROCm, as well as AMD CPUs and AMD Ryzen AI NPUs. Key points:

• Validated backends: ONNX Runtime across CPU (FP32), GPU (MIGraphX Execution Provider — FP32/FP16/BF16/INT8), and NPU (VitisAI Execution Provider — BF16).
• No code changes required versus the upstream Ultralytics YOLO26-cls implementation — only environment/runtime configuration differs.

---

Getting Started

For setup instructions, evaluation scripts, and custom configuration options, see the yolo26_classify on GitHub.

---

Model Details

Model Type: Image classification (single-pass CNN classifier)

Base Model: YOLO26-cls (Ultralytics), size n (nano) default

Model Stats:

• Input (images): (1, 3, 224, 224) float32
• Output (output0): (1, 1000) float32
• Model sizes: n/s/m/l ship pretrained classify weights (x may not resolve to a hosted weight)
• Precision tested: FP32 (CPU); FP32, FP16, BF16, INT8 (GPU); BF16 (NPU)

---

Accuracy Pipeline

Accuracy evaluation is not yet implemented for this model. The bundled imagenette subset (10 ImageNet classes, staged under datasets/samples/) is used only for visual top-5 sanity checks via annotated overlays, not full ImageNet-1000 accuracy scoring.

---

Dig Deeper

Want to explore the full evaluation scripts, config options, and other AMD-optimized model examples?

📂 View the full project on GitHub

The GitHub repository includes:

• Setup and prerequisites for ROCm environments
• Per-operator latency profiling scripts (including NPU AI Analyzer integration)
• Sample-image evaluation with annotated top-5 class overlays
• Benchmarking and reproduction instructions across CPU, GPU, and NPU