Update model card for yolo26-det

README.md

@@ -9,14 +9,11 @@ tags:
   - onnx
   - int8
   - yolo
-  - gstreamer
   - edgefirst
   - nxp
   - hailo
   - jetson
-  - real-time
   - embedded
-  - multiplatform
 model-index:
   - name: yolo26-det
     results:
@@ -28,244 +25,186 @@ model-index:
         metrics:
           - name: "mAP@0.5 (Nano ONNX FP32)"
             type: map_50
-            value: 54.9
+            value: 55.06
           - name: "mAP@0.5-0.95 (Nano ONNX FP32)"
             type: map
-            value: 39.7
-          - name: "mAP@0.5 (Nano TFLite INT8)"
+            value: 39.71
+          - name: "mAP@0.5 (Small ONNX FP32)"
             type: map_50
-            value: 51.5
-          - name: "mAP@0.5-0.95 (Nano TFLite INT8)"
+            value: 63.6
+          - name: "mAP@0.5-0.95 (Small ONNX FP32)"
             type: map
-            value: 34.9
+            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 Edge AI
+# YOLO26 Detection — EdgeFirst Model Zoo
 
-**NXP i.MX 8M Plus** | **NXP i.MX 93** | **NXP i.MX 95** | **NXP Ara240** | **RPi5 + Hailo-8/8L** | **NVIDIA Jetson**
-YOLO26 Detection models optimized for edge AI deployment across multiple hardware platforms. All sizes from Nano to XLarge, in ONNX FP32 and TFLite INT8 formats, with platform-specific compiled models for NPU acceleration.
+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).
 
-Trained on [COCO 2017](https://test.edgefirst.studio/public/projects/2839/home) (80 classes). Part of the [EdgeFirst Model Zoo](https://huggingface.co/spaces/EdgeFirst/Models).
 > [!TIP]
-> **Training session**: [View on EdgeFirst Studio](https://test.edgefirst.studio/public/projects/2839/experiment/training/list?exp_id=4657) — dataset, training config, metrics, and exported artifacts.
+> **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]
-> end2end=False required for INT8. Fastest architecture.
+> End-to-end attention head. `end2end=False` required for INT8 export.
 
 ---
 
-## Size Comparison
+## Reference accuracy — ONNX FP32
 
-All models validated on COCO val2017 (5000 images, 80 classes).
+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 | ONNX FP32 mAP@0.5 | ONNX FP32 mAP@0.5-0.95 | TFLite INT8 mAP@0.5 | TFLite INT8 mAP@0.5-0.95 |
-|------|--------|--------|--------------------|-------------------------|----------------------|--------------------------|
-| Nano | 2.7M | 7.6 | 54.9% | 39.7% | 51.5% | 34.9% |
-| Small | 10.3M | 27.0 | — | — | — | — |
-| Medium | 24.5M | 74.4 | — | — | — | — |
+| 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=metrics) |
+| Small | 10.3M | 27.0 | 63.60% | 47.16% | 51.14% | [v-1d3c](https://test.edgefirst.studio/public/validation/v-1d3c/details?mode=metrics) |
+| Medium | 24.5M | 74.4 | 68.89% | 51.88% | 56.41% | [v-1d3e](https://test.edgefirst.studio/public/validation/v-1d3e/details?mode=metrics) |
 | Large | 42.5M | 155.0 | — | — | — | — |
 | XLarge | 67.5M | 244.0 | — | — | — | — |
 
 ---
 
-## On-Target Performance
-
-Full pipeline timing: pre-processing + inference + post-processing.
-
-| Size | Platform | Pre-proc (ms) | Inference (ms) | Post-proc (ms) | Total (ms) | FPS |
-|------|----------|---------------|----------------|-----------------|------------|-----|
-| — | — | — | — | — | — | — |
-
-*Measured with [EdgeFirst Perception](https://github.com/EdgeFirstAI) stack. Timing includes full GStreamer pipeline overhead.*
-
----
-
-## Downloads
-
-<details open>
-<summary><strong>ONNX FP32</strong> — Any platform with ONNX Runtime.</summary>
-
-| Size | File | Status |
-|------|------|--------|
-| Nano | `yolo26n-det-coco.onnx` | [Download](https://huggingface.co/EdgeFirst/yolo26-det/resolve/main/onnx/yolo26n-det-coco.onnx) |
-| Small | `yolo26s-det-coco.onnx` | [Download](https://huggingface.co/EdgeFirst/yolo26-det/resolve/main/onnx/yolo26s-det-coco.onnx) |
-| Medium | `yolo26m-det-coco.onnx` | [Download](https://huggingface.co/EdgeFirst/yolo26-det/resolve/main/onnx/yolo26m-det-coco.onnx) |
-| Large | `yolo26l-det-coco.onnx` | [Download](https://huggingface.co/EdgeFirst/yolo26-det/resolve/main/onnx/yolo26l-det-coco.onnx) |
-| XLarge | `yolo26x-det-coco.onnx` | [Download](https://huggingface.co/EdgeFirst/yolo26-det/resolve/main/onnx/yolo26x-det-coco.onnx) |
-
-</details>
-
-<details>
-<summary><strong>TFLite INT8</strong> — CPU or NPU via runtime delegate (i.MX 8M Plus VX Delegate).</summary>
-
-| Size | File | Status |
-|------|------|--------|
-| Nano | `yolo26n-det-coco.tflite` | [Download](https://huggingface.co/EdgeFirst/yolo26-det/resolve/main/tflite/yolo26n-det-coco.tflite) |
-| Small | `yolo26s-det-coco.tflite` | [Download](https://huggingface.co/EdgeFirst/yolo26-det/resolve/main/tflite/yolo26s-det-coco.tflite) |
-| Medium | `yolo26m-det-coco.tflite` | [Download](https://huggingface.co/EdgeFirst/yolo26-det/resolve/main/tflite/yolo26m-det-coco.tflite) |
-| Large | `yolo26l-det-coco.tflite` | [Download](https://huggingface.co/EdgeFirst/yolo26-det/resolve/main/tflite/yolo26l-det-coco.tflite) |
-| XLarge | `yolo26x-det-coco.tflite` | [Download](https://huggingface.co/EdgeFirst/yolo26-det/resolve/main/tflite/yolo26x-det-coco.tflite) |
+## On-target validation results
 
-</details>
+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).
 
-<details>
-<summary><strong>NXP i.MX 95 (eIQ Neutron)</strong> — eIQ Neutron NPU optimized.</summary>
-
-| Size | File | Status |
-|------|------|--------|
-| Nano | `yolo26n-det-coco.imx95.tflite` | [Download](https://huggingface.co/EdgeFirst/yolo26-det/resolve/main/imx95/yolo26n-det-coco.imx95.tflite) |
-| Small | `yolo26s-det-coco.imx95.tflite` | Coming Soon |
-| Medium | `yolo26m-det-coco.imx95.tflite` | Coming Soon |
-| Large | `yolo26l-det-coco.imx95.tflite` | Coming Soon |
-| XLarge | `yolo26x-det-coco.imx95.tflite` | Coming Soon |
-
-</details>
+Cells rendered as `—` are sessions that did not meet the EdgeFirst publication threshold; the underlying session is still linked in the Source column for inspection.
 
+| 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=metrics) |
+| Nano | NXP i.MX 95 + eIQ Neutron NPU | — | — | — | — | — | [v-1d57](https://test.edgefirst.studio/public/validation/v-1d57/details?mode=metrics) |
+| 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=metrics) |
+| 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=metrics) |
+| 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=metrics) |
 
 
 ---
 
-## Deploy with EdgeFirst Perception
+## Validation pipeline
 
-Copy-paste [GStreamer](https://github.com/EdgeFirstAI/gstreamer) pipeline examples for each platform.
+These results are produced by the EdgeFirst on-target validation pipeline:
 
-### NXP i.MX 8M Plus — Camera to Detection with Vivante NPU
+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.
 
-```bash
-gst-launch-1.0 \
-  v4l2src device=/dev/video0 ! video/x-raw,width=640,height=480 ! \
-  edgefirstcameraadaptor ! \
-  tensor_filter framework=tensorflow-lite \
-    model=yolo26n-det-coco.tflite \
-    custom=Delegate:External,ExtDelegateLib:libvx_delegate.so ! \
-  edgefirstdetdecoder ! edgefirstoverlay ! waylandsink
-```
+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.
 
-### RPi5 + Hailo-8L
+See [EdgeFirst Studio](https://edgefirst.studio) for the full validation pipeline.
 
-```bash
-gst-launch-1.0 \
-  v4l2src device=/dev/video0 ! video/x-raw,width=640,height=480 ! \
-  hailonet hef-path=yolo26n-det-coco.hailo8l.hef ! \
-  hailofilter function-name=yolo26_nms ! \
-  hailooverlay ! videoconvert ! autovideosink
-```
-
-### NVIDIA Jetson (TensorRT)
+---
 
-```bash
-gst-launch-1.0 \
-  v4l2src device=/dev/video0 ! video/x-raw,width=640,height=480 ! \
-  edgefirstcameraadaptor ! \
-  nvinfer config-file-path=yolo26n-det-coco-config.txt ! \
-  edgefirstdetdecoder ! edgefirstoverlay ! nveglglessink
-```
+## 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.
 
-*Full pipeline documentation: [EdgeFirst GStreamer Plugins](https://github.com/EdgeFirstAI/gstreamer)*
+*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.*
 
 ---
 
-## Foundation (HAL) Python Integration
+## Inference example (Python)
 
 ```python
 from edgefirst.hal import Model, TensorImage
 
-# Load model — metadata (labels, decoder config) is embedded in the file
-model = Model("yolo26n-det-coco.tflite")
+# 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)
 
-# Access detections
+# 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.*
+[EdgeFirst HAL](https://github.com/EdgeFirstAI/hal) — Hardware abstraction layer with accelerated inference delegates.
 
 ---
 
-## CameraAdaptor
-
-EdgeFirst [CameraAdaptor](https://github.com/EdgeFirstAI/cameraadaptor) enables training and inference directly on native sensor formats (GREY, YUYV, etc.) — skipping the ISP color conversion pipeline entirely. This reduces latency and power consumption on edge devices.
+## Traceability
 
-CameraAdaptor variants are included alongside baseline RGB models:
+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:
 
-| Variant | Input Format | Use Case |
-|---------|-------------|----------|
-| `yolo26n-det-coco.onnx` | RGB (3ch) | Standard camera input |
-| `yolo26n-det-coco-grey.onnx` | GREY (1ch) | Monochrome / IR sensors |
-| `yolo26n-det-coco-yuyv.onnx` | YUYV (2ch) | Raw sensor bypass |
+```
+https://test.edgefirst.studio/public/validation/v-XXXX/details?mode=metrics
+```
 
-*Train CameraAdaptor models with [EdgeFirst Studio](https://edgefirst.studio) — the CameraAdaptor layer is automatically inserted during training.*
+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.
 
 ---
 
-## Train Your Own with EdgeFirst Studio
+## See also
 
-Train on your own dataset with [**EdgeFirst Studio**](https://edgefirst.studio):
+Other model families in the [EdgeFirst Model Zoo](https://huggingface.co/spaces/EdgeFirst/Models):
 
-- **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)
+| 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) |
 
 ---
 
-## See Also
+## Train your own with EdgeFirst Studio
 
-Other models in the [EdgeFirst Model Zoo](https://huggingface.co/spaces/EdgeFirst/Models):
+Train on your own dataset with [**EdgeFirst Studio**](https://edgefirst.studio):
 
-| Model | Task | Best Nano Metric | Link |
-|-------|------|-------------------|------|
-| YOLOv5 Detection | Detection | 49.6% mAP@0.5 (ONNX) | [EdgeFirst/yolov5-det](https://huggingface.co/EdgeFirst/yolov5-det) |
-| YOLOv8 Detection | Detection | 50.2% mAP@0.5 (ONNX) | [EdgeFirst/yolov8-det](https://huggingface.co/EdgeFirst/yolov8-det) |
-| YOLOv8 Segmentation | Segmentation | 34.1% Mask mAP@0.5-0.95 (ONNX) | [EdgeFirst/yolov8-seg](https://huggingface.co/EdgeFirst/yolov8-seg) |
-| YOLO11 Detection | Detection | 53.4% mAP@0.5 (ONNX) | [EdgeFirst/yolo11-det](https://huggingface.co/EdgeFirst/yolo11-det) |
-| YOLO11 Segmentation | Segmentation | 35.5% Mask mAP@0.5-0.95 (ONNX) | [EdgeFirst/yolo11-seg](https://huggingface.co/EdgeFirst/yolo11-seg) |
-| YOLO26 Segmentation | Segmentation | 37.0% Mask mAP@0.5-0.95 (ONNX) | [EdgeFirst/yolo26-seg](https://huggingface.co/EdgeFirst/yolo26-seg) |
+- 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 Details
+## Technical notes
+
+### Quantization pipeline
 
-### Quantization Pipeline
+All TFLite INT8 models are produced by EdgeFirst's quantization pipeline ([details](https://github.com/EdgeFirstAI/studio-ultralytics)):
 
-All TFLite INT8 models are produced by EdgeFirst's custom 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
 
-1. **ONNX Export** — Standard Ultralytics export with `simplify=True`
-2. **TF-Wrapped ONNX** — Box coordinates normalized to [0,1] inside DFL decode via `tf_wrapper` (~1.2% better mAP than post-hoc normalization)
-3. **Split Decoder** — Boxes, scores, and mask coefficients split into separate output tensors for independent INT8 quantization scales
-4. **Smart Calibration** — 500 images selected via greedy coverage maximization from COCO val2017
-5. **Full INT8** — `uint8` input (raw pixels), `int8` output (per-tensor scales), MLIR quantizer
+### Split decoder output format
 
-### Split Decoder Output Format
+**Detection** (e.g. yolo26n):
 
-**Detection** (e.g., yolo26n):
-- Boxes: `(1, 4, 8400)` — normalized [0,1] coordinates
-- Scores: `(1, 80, 8400)` — class probabilities
+- `boxes` — `(1, 4, 8400)` normalized `[0, 1]` coordinates
+- `scores` — `(1, 80, 8400)` per-class probabilities
 
-Each tensor has independent quantization scale and zero-point. EdgeFirst HAL handles dequantization and reassembly automatically.
+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.
 
-### Metadata
+### Embedded metadata
 
-- **TFLite**: `edgefirst.json`, `labels.txt`, and `edgefirst.yaml` embedded via ZIP (no `tflite-support` dependency)
-- **ONNX**: `edgefirst.json` embedded via `model.metadata_props`
+- **TFLite**: `edgefirst.json` and `labels.txt` embedded in the ZIP-format model file
+- **ONNX**: `edgefirst.json` embedded in `model.metadata_props`
 
-No standalone metadata files — models are self-contained.
+No sidecar files required; the model artifact is self-contained.
 
 ---
 
 ## Limitations
 
-- **COCO bias** — Models trained on COCO (80 classes) inherit its biases: Western-centric scenes, specific object distributions, limited weather/lighting diversity
-- **INT8 accuracy loss** — Full-integer quantization typically degrades mAP by 6-12% relative to FP32; actual loss depends on model architecture and dataset
-- **Thermal variation** — On-target performance varies with device temperature; sustained inference may throttle on passively-cooled devices
-- **Input resolution** — All models expect 640×640 input; other resolutions require letterboxing or may reduce accuracy
-- **CameraAdaptor variants** — GREY/YUYV models trade color information for latency; accuracy may differ from RGB baseline depending on the task
+- **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.
 
 ---
 
@@ -273,7 +212,7 @@ No standalone metadata files — models are self-contained.
 
 ```bibtex
 @software{edgefirst_yolo26_det,
-  title = { {YOLO26 Detection — EdgeFirst Edge AI} },
+  title = { {YOLO26 Detection — EdgeFirst Model Zoo} },
   author = {Au-Zone Technologies},
   url = {https://huggingface.co/EdgeFirst/yolo26-det},
   year = {2026},