--- license: apache-2.0 library_name: edgefirst pipeline_tag: object-detection tags: - edge-ai - npu - tflite - onnx - int8 - yolo - edgefirst - nxp - hailo - jetson - embedded model-index: - name: yolo26-det results: - task: type: object-detection dataset: name: COCO val2017 type: coco metrics: - name: "mAP@0.5 (Nano ONNX FP32)" type: map_50 value: 55.06 - name: "mAP@0.5-0.95 (Nano ONNX FP32)" type: map value: 39.71 - name: "mAP@0.5 (Small ONNX FP32)" type: map_50 value: 63.6 - name: "mAP@0.5-0.95 (Small ONNX FP32)" type: map value: 47.16 - name: "mAP@0.5 (Medium ONNX FP32)" type: map_50 value: 68.89 - name: "mAP@0.5-0.95 (Medium ONNX FP32)" type: map value: 51.88 --- # YOLO26 Detection — EdgeFirst Model Zoo YOLO26 Detection models trained on [COCO 2017](https://test.edgefirst.studio/public/projects/2839/home) (80 classes) and validated on real edge hardware through the EdgeFirst Profiler + Validator pipeline. Each row in the tables below cites the EdgeFirst Studio validation session (`v-XXXX`) that produced the measurement. Part of the [EdgeFirst Model Zoo](https://huggingface.co/spaces/EdgeFirst/Models). > [!TIP] > **Training experiment**: [View on EdgeFirst Studio](https://test.edgefirst.studio/public/projects/2839/experiment/training/list?exp_id=4657) — dataset, training configuration, metrics, and exported artifacts. > [!NOTE] > End-to-end attention head. `end2end=False` required for INT8 export. --- ## Reference accuracy — ONNX FP32 Accuracy ceiling for each size, measured against COCO `val2017` (5,000 images) with `pycocotools`. Quantized and compiled artifacts (TFLite INT8, HEF, etc.) are graded against this reference per the EdgeFirst publication rule. | Size | Params | GFLOPs | mAP@0.5 | mAP@0.5-0.95 | mAP@0.75 | Source | |------|--------|--------|---------|--------------|----------|--------| | Nano | 2.7M | 7.6 | 55.06% | 39.71% | 42.87% | [v-1d3b](https://test.edgefirst.studio/public/validation/v-1d3b/details?mode=info) | | Small | 10.3M | 27.0 | 63.60% | 47.16% | 51.14% | [v-1d3c](https://test.edgefirst.studio/public/validation/v-1d3c/details?mode=info) | | Medium | 24.5M | 74.4 | 68.89% | 51.88% | 56.41% | [v-1d3e](https://test.edgefirst.studio/public/validation/v-1d3e/details?mode=info) | | Large | 42.5M | 155.0 | — | — | — | — | | XLarge | 67.5M | 244.0 | — | — | — | — | --- ## On-target validation results Each row is one EdgeFirst Studio validation session. Click the **Source** link to inspect the full session — model artifact, dataset version, parameters, per-stage Perfetto trace, and the host hardware description (hostname, kernel version, SoC, NPU, profiler version). Rows without a number under the metric columns are validation sessions that are currently work in progress — for example a larger size that has not yet been profiled on a given NPU, or a (model, NPU) combination where we are tracking a quantization or runtime regression with the toolchain vendor. The underlying Studio session remains linked in the Source column so its current status can be inspected. | Size | Platform | mAP@0.5 | Δ vs FP32 (pp) | mAP@0.5-0.95 | Inference (ms) | FPS (median) | Source | |------|----------|---------|----------------|--------------|----------------|--------------|--------| | Nano | NXP i.MX 8M Plus + VeriSilicon NPU | 50.07% | -4.99 | 32.27% | 105.65 | 8.5 | [v-1d54](https://test.edgefirst.studio/public/validation/v-1d54/details?mode=info) | | Nano | NXP i.MX 95 + eIQ Neutron NPU | — | — | — | — | — | [v-1d57](https://test.edgefirst.studio/public/validation/v-1d57/details?mode=info) | | Nano | Raspberry Pi 5 + Hailo-8L NPU | 51.60% | -3.46 | 35.41% | 23.73 | 41.3 | [v-1d47](https://test.edgefirst.studio/public/validation/v-1d47/details?mode=info) | | Small | Raspberry Pi 5 + Hailo-8L NPU | 60.03% | -3.57 | 42.81% | 48.70 | 20.3 | [v-1d48](https://test.edgefirst.studio/public/validation/v-1d48/details?mode=info) | | Medium | Raspberry Pi 5 + Hailo-8L NPU | 65.64% | -3.25 | 47.16% | 90.55 | 11.0 | [v-1d4d](https://test.edgefirst.studio/public/validation/v-1d4d/details?mode=info) | --- ## Validation pipeline These results are produced by the EdgeFirst on-target validation pipeline: 1. **EdgeFirst Profiler** runs on the target hardware, executes the full inference pipeline (preprocess → inference → postprocess), and emits per-image predictions in EdgeFirst Arrow/Parquet plus a Perfetto trace. 2. **EdgeFirst Validator** consumes the predictions and trace, computes `pycocotools` accuracy metrics and per-stage timing summaries, and publishes the results to the Studio validation session. 3. **EdgeFirst HAL** ([open source](https://github.com/EdgeFirstAI/hal)) provides the hardware-accelerated preprocessing and post-decoding primitives used at both validation and deployment time, so the timings measured here reflect the same accelerated paths a production runtime would take. Inference latency is reported as the on-accelerator inference time. FPS is the measured end-to-end pipelined throughput from the Perfetto trace, which generally exceeds `1000 / (preprocess + inference + postprocess)` because the runtime overlaps stages across frames. See [EdgeFirst Studio](https://edgefirst.studio) for the full validation pipeline. --- ## Downloads Artifacts are organized by deployment target. Each model file embeds the EdgeFirst `edgefirst.json` metadata (training session, dataset version, calibration artifact, converter chain) so a single file is sufficient for deployment — no sidecar configuration required. *Per-artifact download links are populated from the Studio artifact registry. To see the live download table, regenerate this card with `--studio` against an authenticated Studio session.* --- ## Inference example (Python) ```python from edgefirst.hal import Model, TensorImage # Load the model — embedded edgefirst.json carries labels and decoder config model = Model("yolo26n-det-int8.tflite") # Run inference on an image image = TensorImage.from_file("image.jpg") results = model.predict(image) # Iterate detections for det in results.detections: print(f"{det.label}: {det.confidence:.2f} at {det.bbox}") ``` [EdgeFirst HAL](https://github.com/EdgeFirstAI/hal) — Hardware abstraction layer with accelerated inference delegates. --- ## Traceability Every measurement in the tables above is reachable through the EdgeFirst Studio validation framework. The `v-XXXX` Source link on each row resolves to a public Studio URL of the form: ``` https://test.edgefirst.studio/public/validation/v-XXXX/details?mode=info ``` From there, the full provenance chain is one click deeper: training session ID, dataset version, calibration artifact, converter chain (e.g. TFLite quantizer + Neutron compile), validation parameters, and the host hardware description (hostname, kernel version, SoC, NPU, profiler version). The same model file you download from this repository embeds the same chain in its `edgefirst.json` metadata. --- ## See also Other model families in the [EdgeFirst Model Zoo](https://huggingface.co/spaces/EdgeFirst/Models): | Model | Task | Link | |-------|------|------| | YOLOv5 Detection | Detection | [EdgeFirst/yolov5-det](https://huggingface.co/EdgeFirst/yolov5-det) | | YOLOv8 Detection | Detection | [EdgeFirst/yolov8-det](https://huggingface.co/EdgeFirst/yolov8-det) | | YOLOv8 Segmentation | Segmentation | [EdgeFirst/yolov8-seg](https://huggingface.co/EdgeFirst/yolov8-seg) | | YOLO11 Detection | Detection | [EdgeFirst/yolo11-det](https://huggingface.co/EdgeFirst/yolo11-det) | | YOLO11 Segmentation | Segmentation | [EdgeFirst/yolo11-seg](https://huggingface.co/EdgeFirst/yolo11-seg) | | YOLO26 Segmentation | Segmentation | [EdgeFirst/yolo26-seg](https://huggingface.co/EdgeFirst/yolo26-seg) | --- ## Train your own with EdgeFirst Studio Train on your own dataset with [**EdgeFirst Studio**](https://edgefirst.studio): - Free tier includes YOLO training with automatic INT8 quantization and edge deployment. - Upload datasets via [EdgeFirst Recorder](https://github.com/EdgeFirstAI/recorder) or COCO/YOLO format. - AI-assisted annotation with auto-labeling. - CameraAdaptor integration for native sensor format training. - Deploy trained models to edge devices via [EdgeFirst Client](https://github.com/EdgeFirstAI/client). --- ## Technical notes ### Quantization pipeline All TFLite INT8 models are produced by EdgeFirst's quantization pipeline ([details](https://github.com/EdgeFirstAI/studio-ultralytics)): 1. **ONNX export** — standard Ultralytics export with `simplify=True` 2. **TF-wrapped ONNX** — box coordinates normalized to `[0, 1]` inside DFL decode 3. **Split decoder** — boxes, scores, and mask coefficients split into separate output tensors so each receives an independent INT8 quantization scale 4. **Smart calibration** — calibration samples selected via greedy coverage maximization; the artifact is content-addressed by parameter hash and cached in Studio for deterministic reuse 5. **Full integer INT8** — `uint8` input, `int8` output, MLIR quantizer ### Split decoder output format **Detection** (e.g. yolo26n): - `boxes` — `(1, 4, 8400)` normalized `[0, 1]` coordinates - `scores` — `(1, 80, 8400)` per-class probabilities Each tensor has its own quantization scale and zero point. The EdgeFirst HAL handles dequantization and reassembly automatically; no application code change is required across NPU targets. ### Embedded metadata - **TFLite**: `edgefirst.json` and `labels.txt` embedded in the ZIP-format model file - **ONNX**: `edgefirst.json` embedded in `model.metadata_props` No sidecar files required; the model artifact is self-contained. --- ## Limitations - **COCO bias** — models trained on COCO (80 classes) inherit the dataset's biases (Western-centric scenes, particular object distributions, limited weather/lighting diversity). - **INT8 quantization loss** — full-integer quantization introduces accuracy loss relative to FP32; the magnitude per platform is shown in the *Δ vs FP32* column above. - **Input resolution** — all models expect 640×640 input; other resolutions require letterboxing. --- ## Citation ```bibtex @software{edgefirst_yolo26_det, title = { {YOLO26 Detection — EdgeFirst Model Zoo} }, author = {Au-Zone Technologies}, url = {https://huggingface.co/EdgeFirst/yolo26-det}, year = {2026}, license = {Apache-2.0}, } ``` ---

EdgeFirst Studio · GitHub · Docs · Au-Zone Technologies
Apache 2.0 · © Au-Zone Technologies Inc.