Upload folder using huggingface_hub

README.md

@@ -8,107 +8,123 @@ pinned: true
 license: apache-2.0
 ---
 
-# EdgeFirst AI — Spatial Perception at the Edge
+# EdgeFirst Model Zoo
 
-**EdgeFirst Perception** is an open-source suite of libraries and microservices for AI-driven spatial perception on edge devices. It supports cameras, LiDAR, radar, and time-of-flight sensors — enabling real-time object detection, segmentation, sensor fusion, and 3D spatial understanding, optimized for resource-constrained embedded hardware.
+This Space hosts the [EdgeFirst Model Zoo](https://huggingface.co/spaces/EdgeFirst/Models) landing page. The visual interface is rendered from `index.html`.
 
-[![EdgeFirst Studio](https://img.shields.io/badge/EdgeFirst_Studio-3E3371?style=for-the-badge&logoColor=white)](https://edgefirst.studio)
-[![GitHub](https://img.shields.io/badge/GitHub-212529?style=for-the-badge&logo=github&logoColor=white)](https://github.com/EdgeFirstAI)
-[![Documentation](https://img.shields.io/badge/Documentation-1FA0A8?style=for-the-badge&logo=readthedocs&logoColor=white)](https://doc.edgefirst.ai)
-[![Au-Zone Technologies](https://img.shields.io/badge/Au--Zone_Technologies-6C757D?style=for-the-badge)](https://www.au-zone.com)
+Each model family lives in its own HuggingFace repo containing all size variants (nano through x-large) and platform-specific compiled formats. Models are trained and validated on [EdgeFirst Studio](https://edgefirst.studio), then published here.
 
 ---
 
-## Workflow
+## Model Repositories
 
-<img src="https://huggingface.co/spaces/EdgeFirst/README/resolve/main/01-ecosystem.png" alt="EdgeFirst Model Zoo Ecosystem"/>
+### Detection
 
-Every model in the EdgeFirst Model Zoo passes through a validated pipeline. [**EdgeFirst Studio**](https://edgefirst.studio) manages datasets, training, multi-format export (ONNX, TFLite INT8, eIQ Neutron, Kinara DVM, HailoRT HEF, TensorRT), and reference validation. Models are then deployed to our board farm for **full-dataset on-target validation** on real hardware — measuring both accuracy (mAP) and detailed timing breakdown per device. Results are published here on HuggingFace with per-platform performance tables.
+| Repo | Model | Sizes | Nano mAP@0.5 |
+|------|-------|-------|-------------|
+| [EdgeFirst/yolo26-det](https://huggingface.co/EdgeFirst/yolo26-det) | YOLO26 | n/s/m/l/x | 54.9% |
+| [EdgeFirst/yolo11-det](https://huggingface.co/EdgeFirst/yolo11-det) | YOLO11 | n/s/m/l/x | 53.4% |
+| [EdgeFirst/yolov8-det](https://huggingface.co/EdgeFirst/yolov8-det) | YOLOv8 | n/s/m/l/x | 50.2% |
+| [EdgeFirst/yolov5-det](https://huggingface.co/EdgeFirst/yolov5-det) | YOLOv5 | n/s/m/l/x | 49.6% |
 
-## Model Lifecycle
+### Segmentation
 
-<img src="https://huggingface.co/spaces/EdgeFirst/README/resolve/main/02-model-lifecycle.png" alt="Model Lifecycle: Training to Publication"/>
+| Repo | Model | Sizes | Nano Mask mAP |
+|------|-------|-------|--------------|
+| [EdgeFirst/yolo26-seg](https://huggingface.co/EdgeFirst/yolo26-seg) | YOLO26 | n/s/m/l/x | 37.0% |
+| [EdgeFirst/yolo11-seg](https://huggingface.co/EdgeFirst/yolo11-seg) | YOLO11 | n/s/m/l/x | 35.5% |
+| [EdgeFirst/yolov8-seg](https://huggingface.co/EdgeFirst/yolov8-seg) | YOLOv8 | n/s/m/l/x | 34.1% |
 
-## On-Target Validation
+---
 
-<img src="https://huggingface.co/spaces/EdgeFirst/README/resolve/main/03-on-target-validation.png" alt="On-Target Validation Pipeline"/>
+## Repo Structure
+
+Each model repo follows a consistent layout with platform folders:
+
+```
+EdgeFirst/yolov8-det/
+├── README.md                                # Model card
+├── onnx/
+│   ├── yolov8n-det-fp32.onnx
+│   └── ...
+├── tflite/
+│   ├── yolov8n-det-int8.tflite              # Default (logical split-decoder)
+│   ├── yolov8n-det-int8-smart.tflite        # Smart variant
+│   └── ...
+├── imx95/
+│   ├── yolov8n-det-int8.imx95.tflite
+│   └── ...
+├── hailo/
+│   ├── yolov8n-det-int8.hailo8l.hef
+│   └── ...
+└── jetson/
+    ├── yolov8n-det-fp16.orin-nano.engine
+    └── ...
+```
 
-Unlike desktop-only benchmarks, EdgeFirst validates every model on **real target hardware** with the full dataset. Each device produces both accuracy metrics (mAP) and a detailed timing breakdown — load, preprocessing, NPU inference, and decode — so you know exactly how a model performs on your specific platform.
+## Naming Convention
 
----
+**Pattern**: `{version}{size}-{task}-{precision}[-{variant}][.{platform}].{ext}`
 
-## Supported Hardware
+| Component | Description | Examples |
+|-----------|-------------|---------|
+| `{version}{size}` | Model family + variant | `yolov8n`, `yolo11s`, `dfine-n` |
+| `-{task}` | Task suffix | `-det`, `-seg`, `-semseg`, `-depth` |
+| `-{precision}` | Weight precision | `-fp32`, `-fp16`, `-int8` |
+| `-{variant}` | Decoder variant (optional) | `-smart` |
+| `.{platform}` | Deployment target (optional) | `.imx95`, `.ara240`, `.hailo8l`, `.orin-nano` |
+| `.{ext}` | File format | `.onnx`, `.tflite`, `.dvm`, `.hef`, `.engine` |
 
-![NXP i.MX 8M Plus](https://img.shields.io/badge/NXP-i.MX_8M_Plus-3E3371?style=flat-square&logoColor=white)
-![NXP i.MX 95](https://img.shields.io/badge/NXP-i.MX_95-3E3371?style=flat-square&logoColor=white)
-![NXP Ara240](https://img.shields.io/badge/NXP-Ara240-3E3371?style=flat-square&logoColor=white)
-![RPi5 + Hailo-8/8L](https://img.shields.io/badge/RPi5-Hailo--8%2F8L-1FA0A8?style=flat-square&logoColor=white)
-![NVIDIA Jetson](https://img.shields.io/badge/NVIDIA-Jetson-76B900?style=flat-square&logoColor=white)
+**Decoder variants**: No suffix = default for that format (logical split-decoder for INT8, combined for ONNX/float). `-smart` = multi-scale split-decoder offering better accuracy at higher compute cost.
 
----
+**Examples:**
 
-## Model Zoo
+| Description | Filename |
+|-------------|----------|
+| ONNX FP32 (reference) | `yolov8n-det-fp32.onnx` |
+| Generic INT8 TFLite | `yolov8n-det-int8.tflite` |
+| Smart variant TFLite | `yolov8n-det-int8-smart.tflite` |
+| i.MX 95 TFLite | `yolov8n-det-int8.imx95.tflite` |
+| Smart i.MX 95 | `yolov8n-seg-int8-smart.imx95.tflite` |
+| Hailo-8L HEF | `yolov8n-det-int8.hailo8l.hef` |
+| Jetson TensorRT FP16 | `yolov8n-det-fp16.orin-nano.engine` |
 
-Pre-trained YOLO models for edge deployment. Each model repo contains all sizes (nano through x-large), ONNX FP32 and TFLite INT8 formats, with platform-specific compiled variants as they become available.
+## Supported Hardware
 
-### Detection
+- **NXP i.MX 8M Plus** — 2.3 TOPS, TFLite INT8
+- **NXP i.MX 95** — 2.0 TOPS, eIQ Neutron TFLite
+- **NXP Ara240** — 40 eTOPS, Kinara DVM
+- **RPi5 + Hailo-8/8L** — 13–26 TOPS, HailoRT HEF
+- **NVIDIA Jetson Orin** — 67–157 TOPS, TensorRT
 
-| Model | Sizes | Nano mAP@0.5 | Link |
-|-------|-------|-------------|------|
-| **YOLO26** | n/s/m/l/x | 54.9% | [EdgeFirst/yolo26-det](https://huggingface.co/EdgeFirst/yolo26-det) |
-| **YOLO11** | n/s/m/l/x | 53.4% | [EdgeFirst/yolo11-det](https://huggingface.co/EdgeFirst/yolo11-det) |
-| **YOLOv8** | n/s/m/l/x | 50.2% | [EdgeFirst/yolov8-det](https://huggingface.co/EdgeFirst/yolov8-det) |
-| **YOLOv5** | n/s/m/l/x | 49.6% | [EdgeFirst/yolov5-det](https://huggingface.co/EdgeFirst/yolov5-det) |
+## Validation Pipeline
 
-### Instance Segmentation
+Every artifact in the Model Zoo is measured on the same dataset on the same hardware users deploy on. Accuracy numbers and per-stage timing are produced by the same pipeline that runs the deployed model — there is no "benchmark configuration" separate from production.
 
-| Model | Sizes | Nano Mask mAP | Link |
-|-------|-------|--------------|------|
-| **YOLO26** | n/s/m/l/x | 37.0% | [EdgeFirst/yolo26-seg](https://huggingface.co/EdgeFirst/yolo26-seg) |
-| **YOLO11** | n/s/m/l/x | 35.5% | [EdgeFirst/yolo11-seg](https://huggingface.co/EdgeFirst/yolo11-seg) |
-| **YOLOv8** | n/s/m/l/x | 34.1% | [EdgeFirst/yolov8-seg](https://huggingface.co/EdgeFirst/yolov8-seg) |
+### End-to-end flow
 
----
+Each training session produces a single set of weights in [EdgeFirst Studio](https://edgefirst.studio). The export pipeline emits ONNX FP32, INT8 TFLite, and platform-specific compiled formats (i.MX 95 Neutron, NXP Ara-240 DVM, Hailo HEF, Jetson TensorRT). Every output is paired with an on-target validation that captures both accuracy (COCO mAP) and full-pipeline timing. The ONNX FP32 run from each training session is the reference baseline; quantization and runtime loss are measured relative to it.
 
-## Naming Convention
+### EdgeFirst Profiler
 
-| Component | Pattern | Example |
-|-----------|---------|---------|
-| HF Repo | `EdgeFirst/{version}-{task}` | `EdgeFirst/yolov8-det` |
-| ONNX Model | `{version}{size}-{task}.onnx` | `yolov8n-det.onnx` |
-| TFLite Model | `{version}{size}-{task}-int8.tflite` | `yolov8n-det-int8.tflite` |
-| i.MX 95 TFLite | `{version}{size}-{task}.imx95.tflite` | `yolov8n-det.imx95.tflite` |
-| i.MX 93 TFLite | `{version}{size}-{task}.imx93.tflite` | `yolov8n-det.imx93.tflite` |
-| i.MX 943 TFLite | `{version}{size}-{task}.imx943.tflite` | `yolov8n-det.imx943.tflite` |
-| Hailo HEF | `{version}{size}-{task}.hailo{variant}.hef` | `yolov8n-det.hailo8l.hef` |
-| Studio Project | `{Dataset} {Task}` | `COCO Detection` |
-| Studio Experiment | `{Version} {Task}` | `YOLOv8 Detection` |
+The on-target validation agent. Given a model and dataset, it runs full inference on the target device, captures per-image predictions in EdgeFirst Arrow/Parquet, and emits a Perfetto trace alongside. Loads each runtime through its native delegate — VX Delegate on i.MX 8M Plus, eIQ Neutron on i.MX 95, Kinara SDK on Ara-240, HailoRT on RPi5 + Hailo, TensorRT on Jetson — so timing reflects deployed-application reality.
 
-## Validation Pipeline
+### EdgeFirst Validator
+
+The off-target post-processor. Consumes predictions + Perfetto trace, computes the 12-metric COCO accuracy tuple via `pycocotools` (or `lvis-api` for large-vocabulary datasets), and rebuilds per-stage timing summaries from the trace. Results attach to the Studio validation session as a structured YAML payload — the same payload this Model Zoo reads.
 
-| Stage | What | Where |
-|-------|------|-------|
-| **Reference** | ONNX FP32 and TFLite INT8 mAP on full COCO val2017 (5000 images) | EdgeFirst Studio (cloud) |
-| **On-Target** | Full dataset mAP + timing breakdown per device | Board farm (real hardware) |
+### EdgeFirst HAL
 
-## Perception Architecture
+The [EdgeFirst Hardware Abstraction Layer](https://github.com/EdgeFirstAI/hal) provides hardware-accelerated primitives used at both validation and deployment time: letterbox resize, color-space conversion, normalization, layout conversion, YOLO/ModelPack post-decode, NMS. HAL automatically selects DMA-BUF, OpenGL ES, NXP G2D, or CPU paths depending on the platform. Apache 2.0; Rust + Python + C surfaces.
 
-| Layer | Description |
-|-------|-------------|
-| **Foundation** | Hardware abstraction, video I/O, accelerated inference delegates |
-| **Zenoh** | Modular perception pipeline over Zenoh pub/sub |
-| **GStreamer** | Spatial perception elements for GStreamer / NNStreamer |
-| **ROS 2** | Native ROS 2 nodes extending Zenoh microservices *(Roadmap)* |
+### Latency and pipelined throughput
 
-## EdgeFirst Studio
+Two timing surfaces per validation:
 
-[**EdgeFirst Studio**](https://edgefirst.studio) is the MLOps platform that drives the entire model zoo pipeline. **Free tier available.**
+- **`timing.inline`** — per-image `preprocess_ms` / `inference_ms` / `postprocess_ms` with min / mean / median / p95 / p99 / max. The universal contract every producer fills.
+- **`timing.trace`** — full per-stage breakdown from the Perfetto trace (typically 25–33 stages), plus end-to-end FPS distribution.
 
-- Dataset management & AI-assisted annotation
-- Model training with automatic multi-format export and INT8 quantization
-- Reference and on-target validation with full metrics collection
-- CameraAdaptor integration for native sensor format training
-- Deploy trained models to edge devices via the [EdgeFirst Client](https://github.com/EdgeFirstAI/client) CLI
+Throughput exceeds the sum of stage latencies because the runtime pipelines I/O, preprocessing, NPU inference, and decode across frames. The Model Zoo headlines `trace.fps.median` as the throughput number, not the derived `1000 / (preprocess + inference + postprocess)`. Example: YOLOv5n on i.MX 95 Neutron has per-stage means 21.7 + 12.2 + 15.8 ms (naive ≈ 20 FPS) but pipelined throughput of 56 FPS median.
 
 ---
 

index.html

@@ -3,7 +3,7 @@
 <head>
     <meta charset="UTF-8">
     <meta name="viewport" content="width=device-width, initial-scale=1.0">
-    <title>EdgeFirst AI — Spatial Perception at the Edge</title>
+    <title>EdgeFirst AI — Model Zoo</title>
     <base target="_blank">
     <link rel="preconnect" href="https://fonts.googleapis.com">
     <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
@@ -40,6 +40,7 @@
             }
         }
         * { margin: 0; padding: 0; box-sizing: border-box; }
+        html { scroll-behavior: smooth; }
         body {
             font-family: 'Crimson Text', Georgia, serif;
             background: var(--bg);
@@ -47,8 +48,8 @@
             line-height: 1.7;
             font-size: 16px;
         }
-        .container { max-width: 860px; margin: 0 auto; padding: 2.5rem 2rem; }
-        h1, h2, h3 {
+        .container { max-width: 900px; margin: 0 auto; padding: 2.5rem 2rem; }
+        h1, h2, h3, h4 {
             font-family: 'Barlow', -apple-system, sans-serif;
             color: var(--heading);
             line-height: 1.3;
@@ -62,6 +63,16 @@
             margin-top: 2.5rem; margin-bottom: 0.75rem;
             padding-bottom: 0.35rem; border-bottom: 2px solid var(--gold);
         }
+        h3 {
+            font-size: 1.05rem; font-weight: 600;
+            margin-top: 1.75rem; margin-bottom: 0.5rem;
+        }
+        h4 {
+            font-size: 0.9rem; font-weight: 600;
+            color: var(--text-muted);
+            text-transform: uppercase; letter-spacing: 0.04em;
+            margin-top: 1.25rem; margin-bottom: 0.4rem;
+        }
         .tagline {
             font-family: 'Barlow', sans-serif; font-weight: 500;
             letter-spacing: 0.12em; text-transform: uppercase;
@@ -71,72 +82,67 @@
         a { color: var(--link); text-decoration: none; }
         a:hover { text-decoration: underline; color: var(--navy); }
         @media (prefers-color-scheme: dark) { a:hover { color: var(--gold); } }
-        .badges { display: flex; flex-wrap: wrap; gap: 0.4rem; margin: 1.25rem 0; }
-        .badges img { height: 22px; }
-        .link-badges { display: flex; flex-wrap: wrap; gap: 0.4rem; margin: 1.5rem 0; }
-        .link-badges img { height: 28px; }
-        /* Workflow diagram */
-        .diagram-container {
+        /* Navigation */
+        .toc {
+            font-family: 'Barlow', sans-serif;
+            background: var(--bg-subtle);
+            border: 1px solid var(--border);
+            border-radius: 6px;
+            padding: 1rem 1.25rem;
             margin: 1.5rem 0;
-            text-align: center;
+            font-size: 0.88rem;
         }
-        .diagram-container img {
-            max-width: 100%;
-            height: auto;
+        .toc-title {
+            font-weight: 600; font-size: 0.78rem;
+            text-transform: uppercase; letter-spacing: 0.05em;
+            color: var(--text-muted); margin-bottom: 0.5rem;
         }
-        /* Naming table */
-        .naming-table {
-            width: 100%;
-            border-collapse: collapse;
-            margin: 0.75rem 0;
+        .toc ul { list-style: none; display: flex; flex-wrap: wrap; gap: 0.4rem 1.5rem; }
+        .toc a { font-weight: 500; }
+        .breadcrumb {
             font-family: 'Barlow', sans-serif;
-            font-size: 0.9rem;
+            font-size: 0.78rem; color: var(--text-muted);
+            margin-bottom: 0.5rem;
         }
-        .naming-table th {
-            text-align: left;
-            padding: 0.4rem 0.6rem;
-            font-weight: 600;
-            font-size: 0.8rem;
-            text-transform: uppercase;
-            letter-spacing: 0.04em;
-            color: var(--text-muted);
-            border-bottom: 2px solid var(--border);
-        }
-        .naming-table td {
-            padding: 0.35rem 0.6rem;
-            border-bottom: 1px solid var(--border);
-            font-family: 'JetBrains Mono', monospace;
-            font-size: 0.82rem;
-        }
-        .naming-table td:first-child {
+        .breadcrumb a { color: var(--text-muted); font-weight: 500; }
+        .back-to-top {
             font-family: 'Barlow', sans-serif;
-            font-weight: 500;
-            color: var(--text-strong);
+            font-size: 0.75rem; color: var(--text-muted);
+            display: inline-block; margin-top: 0.5rem;
         }
-        /* Architecture table */
-        .arch-table {
-            width: 100%;
-            border-collapse: collapse;
-            margin: 0.75rem 0;
+        /* Badges */
+        .badges { display: flex; flex-wrap: wrap; gap: 0.4rem; margin: 1.25rem 0; }
+        .badges img { height: 22px; }
+        .link-badges { display: flex; flex-wrap: wrap; gap: 0.4rem; margin: 1.5rem 0; }
+        .link-badges img { height: 28px; }
+        .platform-badge {
             font-family: 'Barlow', sans-serif;
-            font-size: 0.95rem;
+            font-size: 0.62rem; font-weight: 600;
+            padding: 0.06rem 0.3rem; border-radius: 2px;
+            letter-spacing: 0.02em; white-space: nowrap;
+            background: var(--bg-card); color: var(--text-muted);
+            border: 1px solid var(--border);
         }
-        .arch-table th {
-            text-align: left;
-            padding: 0.5rem 0.75rem;
-            font-weight: 600; font-size: 0.8rem;
-            text-transform: uppercase; letter-spacing: 0.05em;
-            color: var(--text-muted);
-            border-bottom: 2px solid var(--border);
+        .status-available {
+            font-family: 'Barlow', sans-serif;
+            font-size: 0.68rem; font-weight: 600;
+            padding: 0.1rem 0.4rem; border-radius: 3px;
+            background: #1B7F37; color: #fff;
+            letter-spacing: 0.03em; text-transform: uppercase;
         }
-        .arch-table td {
-            padding: 0.5rem 0.75rem;
-            border-bottom: 1px solid var(--border);
+        .status-coming {
+            font-family: 'Barlow', sans-serif;
+            font-size: 0.68rem; font-weight: 600;
+            padding: 0.1rem 0.4rem; border-radius: 3px;
+            background: var(--gold); color: #333;
+            letter-spacing: 0.03em; text-transform: uppercase;
         }
-        .arch-table td:first-child {
-            font-weight: 600;
-            color: var(--text-strong);
-            white-space: nowrap;
+        .status-planned {
+            font-family: 'Barlow', sans-serif;
+            font-size: 0.68rem; font-weight: 600;
+            padding: 0.1rem 0.4rem; border-radius: 3px;
+            background: var(--indigo); color: #fff;
+            letter-spacing: 0.03em; text-transform: uppercase;
         }
         .roadmap-tag {
             font-family: 'Barlow', sans-serif;
@@ -154,120 +160,93 @@
             vertical-align: middle; margin-left: 0.3rem;
             letter-spacing: 0.03em; text-transform: uppercase;
         }
-        /* Model sub-headings */
-        .container > h3 {
-            font-size: 0.95rem; font-weight: 600;
-            color: var(--text-muted);
-            text-transform: uppercase; letter-spacing: 0.05em;
-            margin-top: 1.25rem; margin-bottom: 0.5rem;
-        }
-        .model-grid {
-            display: grid;
-            grid-template-columns: repeat(auto-fill, minmax(240px, 1fr));
-            gap: 0.75rem; margin: 1rem 0;
-        }
-        .model-card {
-            font-family: 'Barlow', sans-serif;
-            background: var(--bg-card);
-            padding: 0.85rem 1rem; border-radius: 5px;
-            border-left: 3px solid var(--gold);
-            transition: border-color 0.15s;
-        }
-        .model-card:hover { border-left-color: var(--teal); }
-        .model-card h3 { font-size: 0.95rem; font-weight: 600; margin-bottom: 0.2rem; }
-        .model-card h3 a { color: var(--text-strong); }
-        .model-card h3 a:hover { color: var(--link); text-decoration: none; }
-        .model-card .meta { color: var(--text-muted); font-size: 0.82rem; font-weight: 400; }
-        .studio-features { margin: 0.75rem 0 0 1.25rem; color: var(--text); font-size: 0.95rem; }
-        .studio-features li { margin-bottom: 0.25rem; }
-        /* Roadmap */
-        .roadmap-quarter {
-            font-family: 'Barlow', sans-serif;
-            font-size: 0.85rem; font-weight: 600;
-            color: var(--teal-text);
-            text-transform: uppercase; letter-spacing: 0.04em;
-            margin-top: 1.5rem; margin-bottom: 0.4rem;
-            padding-left: 0.6rem;
-            border-left: 3px solid var(--teal);
-        }
-        @media (prefers-color-scheme: dark) {
-            .roadmap-quarter { color: var(--teal); }
-        }
-        .roadmap-table {
+        /* Diagrams */
+        .diagram-container { margin: 1.5rem 0; text-align: center; }
+        .diagram-container img { max-width: 100%; height: auto; }
+        /* Tables */
+        .data-table {
             width: 100%;
             border-collapse: collapse;
-            margin: 0.4rem 0 1.25rem 0;
+            margin: 0.75rem 0;
             font-family: 'Barlow', sans-serif;
             font-size: 0.88rem;
         }
-        .roadmap-table th {
+        .data-table th {
             text-align: left;
-            padding: 0.4rem 0.6rem;
+            padding: 0.5rem 0.6rem;
             font-weight: 600; font-size: 0.75rem;
             text-transform: uppercase; letter-spacing: 0.04em;
             color: var(--text-muted);
             border-bottom: 2px solid var(--border);
         }
-        .roadmap-table td {
-            padding: 0.35rem 0.6rem;
+        .data-table td {
+            padding: 0.4rem 0.6rem;
             border-bottom: 1px solid var(--border);
-            vertical-align: middle;
         }
-        .roadmap-table td:first-child {
+        .data-table td:first-child {
             font-weight: 600;
             color: var(--text-strong);
             white-space: nowrap;
         }
-        .roadmap-table .category {
-            color: var(--text-muted);
-            font-size: 0.8rem;
-        }
-        .badge-row { display: flex; flex-wrap: wrap; gap: 0.25rem; }
-        .license-badge {
-            font-family: 'Barlow', sans-serif;
-            font-size: 0.65rem; font-weight: 600;
-            padding: 0.08rem 0.35rem; border-radius: 3px;
-            letter-spacing: 0.02em; white-space: nowrap;
-        }
-        .license-apache { background: #1B7F37; color: #fff; }
-        .license-mit { background: #0969DA; color: #fff; }
-        .license-agpl { background: #CF222E; color: #fff; }
-        .license-other { background: #6E7781; color: #fff; }
-        .platform-badge {
-            font-family: 'Barlow', sans-serif;
-            font-size: 0.62rem; font-weight: 600;
-            padding: 0.06rem 0.3rem; border-radius: 2px;
-            letter-spacing: 0.02em; white-space: nowrap;
-            background: var(--bg-card); color: var(--text-muted);
-            border: 1px solid var(--border);
+        .data-table code {
+            font-family: 'JetBrains Mono', monospace;
+            font-size: 0.78rem;
         }
-        .status-available {
-            font-family: 'Barlow', sans-serif;
-            font-size: 0.68rem; font-weight: 600;
-            padding: 0.1rem 0.4rem; border-radius: 3px;
-            background: #1B7F37; color: #fff;
-            letter-spacing: 0.03em; text-transform: uppercase;
+        .data-table .num { font-family: 'JetBrains Mono', monospace; font-size: 0.82rem; }
+        .data-table .muted { color: var(--text-muted); }
+        /* Model cards */
+        .model-grid {
+            display: grid;
+            grid-template-columns: repeat(auto-fill, minmax(240px, 1fr));
+            gap: 0.75rem; margin: 1rem 0;
         }
-        .status-coming {
+        .model-card {
             font-family: 'Barlow', sans-serif;
-            font-size: 0.68rem; font-weight: 600;
-            padding: 0.1rem 0.4rem; border-radius: 3px;
-            background: var(--gold); color: #333;
-            letter-spacing: 0.03em; text-transform: uppercase;
+            background: var(--bg-card);
+            padding: 0.85rem 1rem; border-radius: 5px;
+            border-left: 3px solid var(--gold);
+            transition: border-color 0.15s;
         }
-        .status-planned {
+        .model-card:hover { border-left-color: var(--teal); }
+        .model-card h3 { font-size: 0.95rem; font-weight: 600; margin: 0 0 0.2rem 0; }
+        .model-card h3 a { color: var(--text-strong); }
+        .model-card h3 a:hover { color: var(--link); text-decoration: none; }
+        .model-card .meta { color: var(--text-muted); font-size: 0.82rem; font-weight: 400; }
+        /* Accuracy bars */
+        .bar-chart { margin: 0.75rem 0; }
+        .bar-row {
+            display: flex; align-items: center; gap: 0.5rem;
+            font-family: 'Barlow', sans-serif; font-size: 0.82rem;
+            margin-bottom: 0.3rem;
+        }
+        .bar-label { width: 65px; font-weight: 600; color: var(--text-strong); text-align: right; }
+        .bar-track { flex: 1; height: 18px; background: var(--bg-subtle); border-radius: 3px; overflow: hidden; }
+        .bar-fill { height: 100%; border-radius: 3px; display: flex; align-items: center; padding-left: 6px; }
+        .bar-fill span { font-size: 0.72rem; font-weight: 600; color: #fff; white-space: nowrap; }
+        .bar-navy { background: var(--navy); }
+        .bar-teal { background: var(--teal); }
+        .bar-gold { background: var(--gold); }
+        .bar-gold span { color: #333; }
+        /* Collapsible sections */
+        details { margin: 0.5rem 0; }
+        details summary {
             font-family: 'Barlow', sans-serif;
-            font-size: 0.68rem; font-weight: 600;
-            padding: 0.1rem 0.4rem; border-radius: 3px;
-            background: var(--indigo); color: #fff;
-            letter-spacing: 0.03em; text-transform: uppercase;
+            font-weight: 600; font-size: 0.9rem;
+            cursor: pointer; padding: 0.4rem 0;
+            color: var(--text-strong);
         }
+        details summary:hover { color: var(--link); }
+        /* Roadmap */
         .roadmap-note {
             font-family: 'Barlow', sans-serif;
             font-size: 0.82rem; color: var(--text-muted);
-            font-style: italic;
-            margin-top: 0.75rem;
+            font-style: italic; margin-top: 0.75rem;
         }
+        .badge-row { display: flex; flex-wrap: wrap; gap: 0.25rem; }
+        /* Studio features */
+        .studio-features { margin: 0.75rem 0 0 1.25rem; color: var(--text); font-size: 0.95rem; }
+        .studio-features li { margin-bottom: 0.25rem; }
+        /* Footer */
         .footer {
             margin-top: 3rem; padding-top: 1.5rem;
             border-top: 1px solid var(--border);
@@ -277,15 +256,27 @@
         }
         .footer a { color: var(--text-muted); }
         .footer a:hover { color: var(--link); }
+        /* Family section divider */
+        .family-section {
+            margin-top: 2rem;
+            padding-top: 1.5rem;
+            border-top: 1px solid var(--border);
+        }
+        .family-section:first-of-type { border-top: none; padding-top: 0; }
     </style>
 </head>
 <body>
     <div class="container">
-        <h1><span class="edge">Edge</span><span class="first">First</span> AI</h1>
-        <p class="tagline">AI for Spatial Perception</p>
+
+        <!-- ═══════════════════════════════════════════════════════ -->
+        <!-- LEVEL 1: OVERVIEW                                      -->
+        <!-- ═══════════════════════════════════════════════════════ -->
+
+        <h1 id="top"><span class="edge">Edge</span><span class="first">First</span> AI</h1>
+        <p class="tagline">Model Zoo &mdash; Edge AI Perception Models</p>
 
         <p>
-            <strong>EdgeFirst Perception</strong> is an open-source suite of libraries and microservices for AI-driven spatial perception on edge devices. It supports cameras, LiDAR, radar, and time-of-flight sensors &mdash; enabling real-time object detection, segmentation, sensor fusion, and 3D spatial understanding, optimized for resource-constrained embedded hardware.
+            Pre-trained models optimized for edge deployment, validated on real hardware with full-dataset accuracy metrics and per-platform timing breakdowns. Each model repo contains all sizes (nano through x-large) with ONNX FP32, TFLite INT8, and platform-specific compiled formats.
         </p>
 
         <div class="link-badges">
@@ -295,33 +286,35 @@
             <a href="https://www.au-zone.com"><img src="https://img.shields.io/badge/Au--Zone_Technologies-6C757D?style=for-the-badge" alt="Au-Zone Technologies"></a>
         </div>
 
-        <h2>Workflow</h2>
+        <nav class="toc">
+            <div class="toc-title">Model Families</div>
+            <ul>
+                <li><a href="#yolo26" target="_self">YOLO26</a></li>
+                <li><a href="#yolo11" target="_self">YOLO11</a></li>
+                <li><a href="#yolov8" target="_self">YOLOv8</a></li>
+                <li><a href="#yolov5" target="_self">YOLOv5</a></li>
+            </ul>
+        </nav>
 
-        <div class="diagram-container">
-            <img src="01-ecosystem.png" alt="EdgeFirst Model Zoo Ecosystem: Training, Validation, and Publication Workflow">
+        <!-- Overview: All families comparison -->
+        <h2>Detection &mdash; Nano Accuracy Comparison</h2>
+        <p>ONNX FP32 mAP@0.5 on COCO val2017 (5000 images, 80 classes). Nano size for each family.</p>
+        <div class="bar-chart">
+            <div class="bar-row"><span class="bar-label">YOLO26</span><div class="bar-track"><div class="bar-fill bar-navy" style="width:91.5%"><span>54.9%</span></div></div></div>
+            <div class="bar-row"><span class="bar-label">YOLO11</span><div class="bar-track"><div class="bar-fill bar-teal" style="width:89.0%"><span>53.4%</span></div></div></div>
+            <div class="bar-row"><span class="bar-label">YOLOv8</span><div class="bar-track"><div class="bar-fill bar-navy" style="width:83.6%"><span>50.2%</span></div></div></div>
+            <div class="bar-row"><span class="bar-label">YOLOv5</span><div class="bar-track"><div class="bar-fill bar-teal" style="width:82.6%"><span>49.6%</span></div></div></div>
         </div>
 
-        <p>
-            Every model in the EdgeFirst Model Zoo passes through a validated pipeline. <a href="https://edgefirst.studio"><strong>EdgeFirst Studio</strong></a> manages datasets, training, multi-format export (ONNX, TFLite INT8, eIQ Neutron, Kinara DVM, HailoRT HEF, TensorRT), and reference validation. Models are then deployed to our board farm for <strong>full-dataset on-target validation</strong> on real hardware &mdash; measuring both accuracy (mAP) and detailed timing breakdown per device. Results are published here on HuggingFace with per-platform performance tables.
-        </p>
-
-        <h2>Model Lifecycle</h2>
-
-        <div class="diagram-container">
-            <img src="02-model-lifecycle.png" alt="Model Lifecycle: 5 stages from training to publication">
+        <h2>Segmentation &mdash; Nano Accuracy Comparison</h2>
+        <p>ONNX FP32 Mask mAP@0.5-0.95 on COCO val2017. Nano size, split-decoder.</p>
+        <div class="bar-chart">
+            <div class="bar-row"><span class="bar-label">YOLO26</span><div class="bar-track"><div class="bar-fill bar-navy" style="width:92.5%"><span>37.0%</span></div></div></div>
+            <div class="bar-row"><span class="bar-label">YOLO11</span><div class="bar-track"><div class="bar-fill bar-teal" style="width:88.7%"><span>35.5%</span></div></div></div>
+            <div class="bar-row"><span class="bar-label">YOLOv8</span><div class="bar-track"><div class="bar-fill bar-navy" style="width:85.2%"><span>34.1%</span></div></div></div>
         </div>
 
-        <h2>On-Target Validation</h2>
-
-        <div class="diagram-container">
-            <img src="03-on-target-validation.png" alt="On-Target Validation Pipeline: full dataset validation on real hardware">
-        </div>
-
-        <p>
-            Unlike desktop-only benchmarks, EdgeFirst validates every model on <strong>real target hardware</strong> with the full dataset. Each device produces both accuracy metrics (mAP) and a detailed timing breakdown &mdash; load, preprocessing, NPU inference, and decode &mdash; so you know exactly how a model performs on your specific platform.
-        </p>
-
-        <h2>Supported Hardware</h2>
+        <h2>Platform Support</h2>
         <div class="badges">
             <img src="https://img.shields.io/badge/NXP-i.MX_8M_Plus-3E3371?style=flat-square&logoColor=white" alt="NXP i.MX 8M Plus">
             <img src="https://img.shields.io/badge/NXP-i.MX_95-3E3371?style=flat-square&logoColor=white" alt="NXP i.MX 95">
@@ -330,141 +323,327 @@
             <img src="https://img.shields.io/badge/NVIDIA-Jetson-76B900?style=flat-square&logoColor=white" alt="NVIDIA Jetson">
         </div>
 
-        <h2>Model Zoo</h2>
-        <p>Pre-trained YOLO models for edge deployment. Each model repo contains all sizes (nano through x-large), ONNX FP32 and TFLite INT8 formats, with platform-specific compiled variants as they become available.</p>
+        <table class="data-table">
+            <tr><th>Family</th><th>ONNX</th><th>TFLite</th><th>i.MX 93</th><th>i.MX 95</th><th>Ara240</th><th>Hailo</th><th>Jetson</th></tr>
+            <tr><td>YOLO26</td><td>✓</td><td>✓</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+            <tr><td>YOLO11</td><td>✓</td><td>✓</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+            <tr><td>YOLOv8</td><td>✓</td><td>✓</td><td class="muted">&mdash;</td><td><span class="wip-tag">WIP</span></td><td class="muted">&mdash;</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+            <tr><td>YOLOv5</td><td>✓</td><td>✓</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+        </table>
+
+        <!-- ═══════════════════════════════════════════════════════ -->
+        <!-- LEVEL 2: MODEL FAMILIES                                -->
+        <!-- ═══════════════════════════════════════════════════════ -->
+
+        <!-- ── YOLO26 ──────────────────────────────────────────── -->
+        <div class="family-section" id="yolo26">
+            <p class="breadcrumb"><a href="#top" target="_self">Models</a> &rsaquo; <strong>YOLO26</strong></p>
+            <h2>YOLO26</h2>
+            <p>YOLO architecture with end-to-end attention head. <em>Note: <code>end2end=False</code> required for INT8 export.</em></p>
+
+            <h4>Tasks</h4>
+            <div class="model-grid">
+                <div class="model-card">
+                    <h3><a href="#yolo26-det" target="_self">Detection</a></h3>
+                    <p class="meta">n/s/m/l/x &middot; Nano mAP@0.5: 54.9% &middot; <a href="https://huggingface.co/EdgeFirst/yolo26-det">HF Repo</a></p>
+                </div>
+                <div class="model-card">
+                    <h3><a href="#yolo26-seg" target="_self">Segmentation</a></h3>
+                    <p class="meta">n/s/m/l/x &middot; Nano Mask mAP: 37.0% &middot; <a href="https://huggingface.co/EdgeFirst/yolo26-seg">HF Repo</a></p>
+                </div>
+            </div>
+
+            <h4>Size Scaling &mdash; Detection (ONNX FP32)</h4>
+            <table class="data-table">
+                <tr><th>Size</th><th>Params</th><th>GFLOPs</th><th>mAP@0.5</th><th>mAP@0.5-0.95</th></tr>
+                <tr><td>Nano</td><td class="num">2.7M</td><td class="num">7.6</td><td class="num">54.9%</td><td class="num">39.7%</td></tr>
+                <tr><td>Small</td><td class="num">10.3M</td><td class="num">27.0</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+                <tr><td>Medium</td><td class="num">24.5M</td><td class="num">74.4</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+                <tr><td>Large</td><td class="num">42.5M</td><td class="num">155.0</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+                <tr><td>XLarge</td><td class="num">67.5M</td><td class="num">244.0</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+            </table>
+
+            <!-- Level 3: YOLO26 Detection -->
+            <div id="yolo26-det">
+                <p class="breadcrumb"><a href="#top" target="_self">Models</a> &rsaquo; <a href="#yolo26" target="_self">YOLO26</a> &rsaquo; <strong>Detection</strong></p>
+                <h3>YOLO26 Detection</h3>
+                <p>Accuracy on COCO val2017 (5000 images, 80 classes). <a href="https://huggingface.co/EdgeFirst/yolo26-det">View on HuggingFace &rarr;</a></p>
+                <table class="data-table">
+                    <tr><th>Size</th><th>ONNX FP32 mAP@0.5</th><th>TFLite INT8 mAP@0.5</th><th>INT8 Drop</th></tr>
+                    <tr><td>Nano</td><td class="num">54.9%</td><td class="num">51.5%</td><td class="num">−3.4 pp</td></tr>
+                </table>
+            </div>
 
-        <h3>Detection</h3>
-        <div class="model-grid">
-            <div class="model-card">
-                <h3><a href="https://huggingface.co/EdgeFirst/yolo26-det">YOLO26</a></h3>
-                <p class="meta">n/s/m/l/x &middot; COCO 80 classes &middot; Nano mAP@0.5: 54.9%</p>
+            <!-- Level 3: YOLO26 Segmentation -->
+            <div id="yolo26-seg">
+                <p class="breadcrumb"><a href="#top" target="_self">Models</a> &rsaquo; <a href="#yolo26" target="_self">YOLO26</a> &rsaquo; <strong>Segmentation</strong></p>
+                <h3>YOLO26 Segmentation</h3>
+                <p>Split-decoder architecture. <a href="https://huggingface.co/EdgeFirst/yolo26-seg">View on HuggingFace &rarr;</a></p>
+                <table class="data-table">
+                    <tr><th>Size</th><th>ONNX Det mAP</th><th>INT8 Det mAP</th><th>ONNX Mask mAP</th><th>INT8 Mask mAP</th></tr>
+                    <tr><td>Nano</td><td class="num">29.6%</td><td class="num">26.8%</td><td class="num">37.0%</td><td class="num">34.5%</td></tr>
+                </table>
             </div>
-            <div class="model-card">
-                <h3><a href="https://huggingface.co/EdgeFirst/yolo11-det">YOLO11</a></h3>
-                <p class="meta">n/s/m/l/x &middot; COCO 80 classes &middot; Nano mAP@0.5: 53.4%</p>
+            <a class="back-to-top" href="#top" target="_self">&uarr; Back to top</a>
+        </div>
+
+        <!-- ── YOLO11 ──────────────────────────────────────────── -->
+        <div class="family-section" id="yolo11">
+            <p class="breadcrumb"><a href="#top" target="_self">Models</a> &rsaquo; <strong>YOLO11</strong></p>
+            <h2>YOLO11</h2>
+            <p>Newer architecture with attention blocks. Strong balance of accuracy and efficiency.</p>
+
+            <h4>Tasks</h4>
+            <div class="model-grid">
+                <div class="model-card">
+                    <h3><a href="#yolo11-det" target="_self">Detection</a></h3>
+                    <p class="meta">n/s/m/l/x &middot; Nano mAP@0.5: 53.4% &middot; <a href="https://huggingface.co/EdgeFirst/yolo11-det">HF Repo</a></p>
+                </div>
+                <div class="model-card">
+                    <h3><a href="#yolo11-seg" target="_self">Segmentation</a></h3>
+                    <p class="meta">n/s/m/l/x &middot; Nano Mask mAP: 35.5% &middot; <a href="https://huggingface.co/EdgeFirst/yolo11-seg">HF Repo</a></p>
+                </div>
             </div>
-            <div class="model-card">
-                <h3><a href="https://huggingface.co/EdgeFirst/yolov8-det">YOLOv8</a></h3>
-                <p class="meta">n/s/m/l/x &middot; COCO 80 classes &middot; Nano mAP@0.5: 50.2%</p>
+
+            <h4>Size Scaling &mdash; Detection (ONNX FP32)</h4>
+            <table class="data-table">
+                <tr><th>Size</th><th>Params</th><th>GFLOPs</th><th>mAP@0.5</th><th>mAP@0.5-0.95</th></tr>
+                <tr><td>Nano</td><td class="num">2.6M</td><td class="num">6.5</td><td class="num">53.4%</td><td class="num">37.9%</td></tr>
+                <tr><td>Small</td><td class="num">9.4M</td><td class="num">21.5</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+                <tr><td>Medium</td><td class="num">20.1M</td><td class="num">68.0</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+                <tr><td>Large</td><td class="num">25.3M</td><td class="num">87.6</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+                <tr><td>XLarge</td><td class="num">56.9M</td><td class="num">195.0</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+            </table>
+
+            <div id="yolo11-det">
+                <p class="breadcrumb"><a href="#top" target="_self">Models</a> &rsaquo; <a href="#yolo11" target="_self">YOLO11</a> &rsaquo; <strong>Detection</strong></p>
+                <h3>YOLO11 Detection</h3>
+                <p><a href="https://huggingface.co/EdgeFirst/yolo11-det">View on HuggingFace &rarr;</a></p>
+                <table class="data-table">
+                    <tr><th>Size</th><th>ONNX FP32 mAP@0.5</th><th>TFLite INT8 mAP@0.5</th><th>INT8 Drop</th></tr>
+                    <tr><td>Nano</td><td class="num">53.4%</td><td class="num">50.1%</td><td class="num">−3.3 pp</td></tr>
+                </table>
             </div>
-            <div class="model-card">
-                <h3><a href="https://huggingface.co/EdgeFirst/yolov5-det">YOLOv5</a></h3>
-                <p class="meta">n/s/m/l/x &middot; COCO 80 classes &middot; Nano mAP@0.5: 49.6%</p>
+
+            <div id="yolo11-seg">
+                <p class="breadcrumb"><a href="#top" target="_self">Models</a> &rsaquo; <a href="#yolo11" target="_self">YOLO11</a> &rsaquo; <strong>Segmentation</strong></p>
+                <h3>YOLO11 Segmentation</h3>
+                <p><a href="https://huggingface.co/EdgeFirst/yolo11-seg">View on HuggingFace &rarr;</a></p>
+                <table class="data-table">
+                    <tr><th>Size</th><th>ONNX Det mAP</th><th>INT8 Det mAP</th><th>ONNX Mask mAP</th><th>INT8 Mask mAP</th></tr>
+                    <tr><td>Nano</td><td class="num">28.4%</td><td class="num">27.1%</td><td class="num">35.5%</td><td class="num">34.4%</td></tr>
+                </table>
             </div>
+            <a class="back-to-top" href="#top" target="_self">&uarr; Back to top</a>
         </div>
 
-        <h3>Instance Segmentation</h3>
-        <div class="model-grid">
-            <div class="model-card">
-                <h3><a href="https://huggingface.co/EdgeFirst/yolo26-seg">YOLO26</a></h3>
-                <p class="meta">n/s/m/l/x &middot; COCO 80 classes &middot; Nano Mask mAP: 37.0%</p>
+        <!-- ── YOLOv8 ──────────────────────────────────────────── -->
+        <div class="family-section" id="yolov8">
+            <p class="breadcrumb"><a href="#top" target="_self">Models</a> &rsaquo; <strong>YOLOv8</strong></p>
+            <h2>YOLOv8</h2>
+            <p>Anchor-free DFL detection head. Available in detection and instance-segmentation variants.</p>
+
+            <h4>Tasks</h4>
+            <div class="model-grid">
+                <div class="model-card">
+                    <h3><a href="#yolov8-det" target="_self">Detection</a></h3>
+                    <p class="meta">n/s/m/l/x &middot; Nano mAP@0.5: 50.2% &middot; <a href="https://huggingface.co/EdgeFirst/yolov8-det">HF Repo</a></p>
+                </div>
+                <div class="model-card">
+                    <h3><a href="#yolov8-seg" target="_self">Segmentation</a></h3>
+                    <p class="meta">n/s/m/l/x &middot; Nano Mask mAP: 34.1% &middot; <a href="https://huggingface.co/EdgeFirst/yolov8-seg">HF Repo</a></p>
+                </div>
             </div>
-            <div class="model-card">
-                <h3><a href="https://huggingface.co/EdgeFirst/yolo11-seg">YOLO11</a></h3>
-                <p class="meta">n/s/m/l/x &middot; COCO 80 classes &middot; Nano Mask mAP: 35.5%</p>
+
+            <h4>Size Scaling &mdash; Detection (ONNX FP32)</h4>
+            <table class="data-table">
+                <tr><th>Size</th><th>Params</th><th>GFLOPs</th><th>mAP@0.5</th><th>mAP@0.5-0.95</th></tr>
+                <tr><td>Nano</td><td class="num">3.2M</td><td class="num">8.9</td><td class="num">50.2%</td><td class="num">35.8%</td></tr>
+                <tr><td>Small</td><td class="num">11.2M</td><td class="num">28.8</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+                <tr><td>Medium</td><td class="num">25.9M</td><td class="num">79.3</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+                <tr><td>Large</td><td class="num">43.7M</td><td class="num">165.7</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+                <tr><td>XLarge</td><td class="num">68.2M</td><td class="num">258.5</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+            </table>
+
+            <div id="yolov8-det">
+                <p class="breadcrumb"><a href="#top" target="_self">Models</a> &rsaquo; <a href="#yolov8" target="_self">YOLOv8</a> &rsaquo; <strong>Detection</strong></p>
+                <h3>YOLOv8 Detection</h3>
+                <p><a href="https://huggingface.co/EdgeFirst/yolov8-det">View on HuggingFace &rarr;</a></p>
+                <table class="data-table">
+                    <tr><th>Size</th><th>ONNX FP32 mAP@0.5</th><th>TFLite INT8 mAP@0.5</th><th>INT8 Drop</th></tr>
+                    <tr><td>Nano</td><td class="num">50.2%</td><td class="num">47.3%</td><td class="num">−2.9 pp</td></tr>
+                </table>
             </div>
-            <div class="model-card">
-                <h3><a href="https://huggingface.co/EdgeFirst/yolov8-seg">YOLOv8</a></h3>
-                <p class="meta">n/s/m/l/x &middot; COCO 80 classes &middot; Nano Mask mAP: 34.1%</p>
+
+            <div id="yolov8-seg">
+                <p class="breadcrumb"><a href="#top" target="_self">Models</a> &rsaquo; <a href="#yolov8" target="_self">YOLOv8</a> &rsaquo; <strong>Segmentation</strong></p>
+                <h3>YOLOv8 Segmentation</h3>
+                <p><a href="https://huggingface.co/EdgeFirst/yolov8-seg">View on HuggingFace &rarr;</a></p>
+                <table class="data-table">
+                    <tr><th>Size</th><th>ONNX Det mAP</th><th>INT8 Det mAP</th><th>ONNX Mask mAP</th><th>INT8 Mask mAP</th></tr>
+                    <tr><td>Nano</td><td class="num">26.7%</td><td class="num">26.0%</td><td class="num">34.1%</td><td class="num">33.5%</td></tr>
+                </table>
             </div>
+            <a class="back-to-top" href="#top" target="_self">&uarr; Back to top</a>
         </div>
 
-        <h2>Roadmap</h2>
-        <p>The EdgeFirst Model Zoo is expanding across the full spatial perception stack &mdash; from 2D detection through depth estimation, 3D scene understanding, and edge VLMs. All models are validated on real hardware with the same pipeline used for our YOLO models.</p>
+        <!-- ── YOLOv5 ──────────────────────────────────────────── -->
+        <div class="family-section" id="yolov5">
+            <p class="breadcrumb"><a href="#top" target="_self">Models</a> &rsaquo; <strong>YOLOv5</strong></p>
+            <h2>YOLOv5</h2>
+            <p>CSP-Darknet backbone with anchor-based detection head. Detection only.</p>
+
+            <h4>Tasks</h4>
+            <div class="model-grid">
+                <div class="model-card">
+                    <h3><a href="#yolov5-det" target="_self">Detection</a></h3>
+                    <p class="meta">n/s/m/l/x &middot; Nano mAP@0.5: 49.6% &middot; <a href="https://huggingface.co/EdgeFirst/yolov5-det">HF Repo</a></p>
+                </div>
+            </div>
+
+            <h4>Size Scaling &mdash; Detection (ONNX FP32)</h4>
+            <table class="data-table">
+                <tr><th>Size</th><th>Params</th><th>GFLOPs</th><th>mAP@0.5</th><th>mAP@0.5-0.95</th></tr>
+                <tr><td>Nano</td><td class="num">1.9M</td><td class="num">4.5</td><td class="num">49.6%</td><td class="num">33.0%</td></tr>
+                <tr><td>Small</td><td class="num">7.2M</td><td class="num">16.5</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+                <tr><td>Medium</td><td class="num">21.2M</td><td class="num">49.0</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+                <tr><td>Large</td><td class="num">46.5M</td><td class="num">109.1</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+                <tr><td>XLarge</td><td class="num">86.7M</td><td class="num">205.7</td><td class="muted">&mdash;</td><td class="muted">&mdash;</td></tr>
+            </table>
+
+            <div id="yolov5-det">
+                <p class="breadcrumb"><a href="#top" target="_self">Models</a> &rsaquo; <a href="#yolov5" target="_self">YOLOv5</a> &rsaquo; <strong>Detection</strong></p>
+                <h3>YOLOv5 Detection</h3>
+                <p><a href="https://huggingface.co/EdgeFirst/yolov5-det">View on HuggingFace &rarr;</a></p>
+                <table class="data-table">
+                    <tr><th>Size</th><th>ONNX FP32 mAP@0.5</th><th>TFLite INT8 mAP@0.5</th><th>INT8 Drop</th></tr>
+                    <tr><td>Nano</td><td class="num">49.6%</td><td class="num">46.2%</td><td class="num">−3.4 pp</td></tr>
+                </table>
+            </div>
+            <a class="back-to-top" href="#top" target="_self">&uarr; Back to top</a>
+        </div>
+
+        <!-- ═══════════════════════════════════════════════════════ -->
+        <!-- WORKFLOW & VALIDATION                                  -->
+        <!-- ═══════════════════════════════════════════════════════ -->
+
+        <h2>Validation Pipeline</h2>
+        <div class="diagram-container">
+            <img src="01-ecosystem.png" alt="EdgeFirst Model Zoo Ecosystem">
+        </div>
+        <p>
+            Every artifact in the Model Zoo is measured on the <strong>same dataset on the same hardware</strong> users deploy on. <a href="https://edgefirst.studio"><strong>EdgeFirst Studio</strong></a> manages datasets, training, multi-format export, and reference validation; on-target runs happen on a board farm of i.MX 8M Plus, i.MX 95, Ara-240, Hailo, and Jetson devices. Accuracy numbers and per-stage timing are pushed back to Studio session metrics and consumed by this Model Zoo when generating each model card.
+        </p>
+
+        <h3>End-to-end Flow</h3>
+        <div class="diagram-container">
+            <img src="02-model-lifecycle.png" alt="Model Lifecycle: 5 stages from training to publication">
+        </div>
+        <p>
+            Each training session produces a single set of weights. The export pipeline emits ONNX FP32, INT8 TFLite, and platform-specific compiled formats (i.MX 95 Neutron, NXP Ara-240 DVM, Hailo HEF, Jetson TensorRT). Every output is paired with an on-target validation run that captures both <strong>accuracy</strong> (COCO/LVIS mAP against the validation set) and <strong>full-pipeline timing</strong>. The ONNX FP32 run from each training session serves as the reference baseline; quantization and runtime loss are measured relative to it, not relative to externally-published numbers.
+        </p>
 
-        <table class="roadmap-table">
+        <h3>EdgeFirst Profiler</h3>
+        <div class="diagram-container">
+            <img src="03-on-target-validation.png" alt="On-Target Validation Pipeline">
+        </div>
+        <p>
+            The <strong>EdgeFirst Profiler</strong> is the on-target agent that drives every validation. Given a model and a dataset, it runs full inference on the target device, captures per-image predictions in the EdgeFirst Arrow/Parquet format, and emits a <a href="https://perfetto.dev"><strong>Perfetto</strong></a> trace alongside the predictions. The Profiler is hardware-aware: it loads each runtime through its native delegate &mdash; Verisilicon VX Delegate on i.MX 8M Plus, eIQ Neutron Delegate on i.MX 95, Kinara SDK on Ara-240, HailoRT on RPi5 + Hailo, TensorRT on Jetson &mdash; so every stage timed by the trace corresponds to what a deployed application would experience.
+        </p>
+
+        <h3>EdgeFirst Validator</h3>
+        <p>
+            The <strong>EdgeFirst Validator</strong> is the off-target post-processor. It consumes the Profiler's predictions and Perfetto trace, computes the full 12-metric COCO accuracy tuple via <code>pycocotools</code> (or <code>lvis-api</code> for large-vocabulary datasets), and rebuilds per-stage timing summaries from the trace. Results land in a structured YAML payload attached to the Studio validation session &mdash; the same payload the Model Zoo reads to render this page. Accuracy and timing are computed independently of the runtime that produced the predictions, so toolchain regressions surface as cross-platform divergence rather than silent failures.
+        </p>
+
+        <h3>EdgeFirst HAL</h3>
+        <p>
+            The <a href="https://github.com/EdgeFirstAI/hal"><strong>EdgeFirst Hardware Abstraction Layer</strong> (HAL)</a> provides the hardware-accelerated primitives used at both validation and deployment time. The Profiler uses HAL for letterbox resize, color-space conversion, normalization, layout conversion, and post-decode (YOLO/ModelPack output decoding, NMS, mask materialisation). HAL automatically selects DMA-BUF, OpenGL ES, NXP G2D, or CPU paths depending on the platform &mdash; so the timing measured during validation reflects the same accelerated path a production runtime would take. HAL ships as a Rust crate, a Python package, and a C library under Apache 2.0.
+        </p>
+
+        <h3>Latency and Pipelined Throughput</h3>
+        <p>
+            Two complementary timing surfaces are reported per validation:
+        </p>
+        <table class="data-table">
+            <tr><th>Surface</th><th>What it captures</th><th>When it&#39;s present</th></tr>
+            <tr><td><code>timing.inline</code></td><td>Per-image <code>preprocess_ms</code>, <code>inference_ms</code>, <code>postprocess_ms</code> with min / mean / median / p95 / p99 / max</td><td>Always &mdash; the universal contract every producer fills in</td></tr>
+            <tr><td><code>timing.trace</code></td><td>Full per-stage breakdown from the Perfetto trace (typically 25&ndash;33 stages including delegate work, tensor moves, decode passes, NMS, parquet flush), plus end-to-end FPS distribution</td><td>When the Profiler emits a sidecar trace (almost all runs)</td></tr>
+        </table>
+        <p>
+            <strong>Throughput exceeds the sum of stage latencies</strong> because the runtime pipelines I/O, host preprocessing, NPU inference, and decode across frames. Reporting <code>1000 / (preprocess + inference + postprocess)</code> understates real throughput; the Model Zoo uses the <strong>measured end-to-end FPS</strong> from the trace (<code>trace.fps.median</code>) as the headline number. As a concrete example, YOLOv5n on i.MX 95 Neutron has per-stage means 21.7 ms preprocess + 12.2 ms inference + 15.8 ms postprocess (naive estimate ~20 FPS), but the measured pipelined throughput is <strong>56 FPS median</strong> &mdash; the 2.8&times; gap is the value the pipelining delivers.
+        </p>
+
+        <!-- ═══════════════════════════════════════════════════════ -->
+        <!-- NAMING & VARIANTS                                      -->
+        <!-- ═══════════════════════════════════════════════════════ -->
+
+        <h2>Naming Convention</h2>
+        <p>Each HuggingFace repo contains one model family for one task, organized by platform folders.</p>
+        <table class="data-table">
+            <tr><th>Component</th><th>Pattern</th><th>Example</th></tr>
+            <tr><td>HF Repo</td><td><code>EdgeFirst/{version}-{task}</code></td><td><code>EdgeFirst/yolov8-det</code></td></tr>
+            <tr><td>ONNX</td><td><code>{ver}{sz}-{task}-{prec}.onnx</code></td><td><code>yolov8n-det-fp32.onnx</code></td></tr>
+            <tr><td>TFLite</td><td><code>{ver}{sz}-{task}-{prec}.tflite</code></td><td><code>yolov8n-det-int8.tflite</code></td></tr>
+            <tr><td>Smart</td><td><code>{ver}{sz}-{task}-{prec}-smart.tflite</code></td><td><code>yolov8n-seg-int8-smart.tflite</code></td></tr>
+            <tr><td>i.MX 95</td><td><code>{ver}{sz}-{task}-{prec}.imx95.tflite</code></td><td><code>yolov8n-det-int8.imx95.tflite</code></td></tr>
+            <tr><td>Hailo HEF</td><td><code>{ver}{sz}-{task}-{prec}.hailo8l.hef</code></td><td><code>yolov8n-det-int8.hailo8l.hef</code></td></tr>
+            <tr><td>TensorRT</td><td><code>{ver}{sz}-{task}-{prec}.{gpu}.engine</code></td><td><code>yolov8n-det-fp16.orin-nano.engine</code></td></tr>
+        </table>
+
+        <h3>Decoder Variants</h3>
+        <p>INT8 quantized models are available in two decoder configurations. The <strong>default</strong> (no suffix) uses a logical split-decoder &mdash; lossless and zero additional cost. The <strong>-smart</strong> variant uses a multi-scale split-decoder for improved accuracy at the cost of more CPU post-processing. Both are published so users can choose the tradeoff.</p>
+
+        <!-- ═══════════════════════════════════════════════════════ -->
+        <!-- ROADMAP                                                -->
+        <!-- ═══════════════════════════════════════════════════════ -->
+
+        <h2>Roadmap</h2>
+        <p>Expanding across the full spatial perception stack. All models validated on real hardware.</p>
+        <table class="data-table">
             <tr><th>Category</th><th>Examples</th><th>Platforms</th><th>Status</th></tr>
             <tr>
-                <td>Detection (Apache 2.0)</td>
-                <td class="category">DETR-class, EfficientDet, mobile-optimized detectors</td>
+                <td>Detection</td>
+                <td>DETR-class, EfficientDet, mobile-optimized</td>
                 <td><div class="badge-row"><span class="platform-badge">i.MX</span> <span class="platform-badge">Ara240</span> <span class="platform-badge">Hailo</span> <span class="platform-badge">Jetson</span></div></td>
                 <td><span class="status-coming">Coming Soon</span></td>
             </tr>
             <tr>
-                <td>Semantic Segmentation</td>
-                <td class="category">Lightweight real-time scene parsing</td>
+                <td>Semantic Seg</td>
+                <td>Lightweight real-time scene parsing</td>
                 <td><div class="badge-row"><span class="platform-badge">i.MX</span> <span class="platform-badge">Ara240</span> <span class="platform-badge">Hailo</span> <span class="platform-badge">Jetson</span></div></td>
                 <td><span class="status-planned">Roadmap</span></td>
             </tr>
             <tr>
-                <td>Instance Segmentation (Apache 2.0)</td>
-                <td class="category">Non-YOLO mask prediction</td>
-                <td><div class="badge-row"><span class="platform-badge">Ara240</span> <span class="platform-badge">Jetson</span></div></td>
-                <td><span class="status-planned">Roadmap</span></td>
-            </tr>
-            <tr>
-                <td>SAM-like Segmentation</td>
-                <td class="category">Prompted, class-agnostic masks</td>
+                <td>SAM Seg</td>
+                <td>Prompted, class-agnostic masks</td>
                 <td><div class="badge-row"><span class="platform-badge">Ara240</span> <span class="platform-badge">Jetson</span></div></td>
                 <td><span class="status-planned">Roadmap</span></td>
             </tr>
             <tr>
                 <td>Monocular Depth</td>
-                <td class="category">Relative and metric depth estimation</td>
+                <td>Relative and metric depth estimation</td>
                 <td><div class="badge-row"><span class="platform-badge">i.MX</span> <span class="platform-badge">Ara240</span> <span class="platform-badge">Jetson</span></div></td>
                 <td><span class="status-planned">Roadmap</span></td>
             </tr>
             <tr>
-                <td>Stereo Depth</td>
-                <td class="category">Hardware stereo depth matching</td>
-                <td><div class="badge-row"><span class="platform-badge">Jetson</span></div></td>
-                <td><span class="status-planned">Roadmap</span></td>
-            </tr>
-            <tr>
-                <td>3D Detection &amp; Occupancy</td>
-                <td class="category">Monocular 3D, BEV, occupancy grids</td>
+                <td>3D &amp; Occupancy</td>
+                <td>Monocular 3D, BEV, occupancy grids</td>
                 <td><div class="badge-row"><span class="platform-badge">Jetson</span></div></td>
                 <td><span class="status-planned">Roadmap</span></td>
             </tr>
             <tr>
                 <td>Edge VLMs</td>
-                <td class="category">Visual language models for edge inference</td>
+                <td>Visual language models for edge</td>
                 <td><div class="badge-row"><span class="platform-badge">Ara240</span> <span class="platform-badge">Jetson</span></div></td>
                 <td><span class="status-planned">Roadmap</span></td>
             </tr>
         </table>
-
         <p class="roadmap-note">Roadmap is subject to change. Models are published as validation completes on each target platform.</p>
 
-        <h2>Naming Convention</h2>
-        <p>Each HuggingFace repo contains one model family for one task, with all size variants inside.</p>
-        <table class="naming-table">
-            <tr><th>Component</th><th>Pattern</th><th>Example</th></tr>
-            <tr><td>HF Repo</td><td>EdgeFirst/{version}-{task}</td><td>EdgeFirst/yolov8-det</td></tr>
-            <tr><td>ONNX Model</td><td>{version}{size}-{task}.onnx</td><td>yolov8n-det.onnx</td></tr>
-            <tr><td>TFLite Model</td><td>{version}{size}-{task}-int8.tflite</td><td>yolov8n-det-int8.tflite</td></tr>
-            <tr><td>i.MX 95 TFLite</td><td>{version}{size}-{task}.imx95.tflite</td><td>yolov8n-det.imx95.tflite</td></tr>
-            <tr><td>i.MX 93 TFLite</td><td>{version}{size}-{task}.imx93.tflite</td><td>yolov8n-det.imx93.tflite</td></tr>
-            <tr><td>i.MX 943 TFLite</td><td>{version}{size}-{task}.imx943.tflite</td><td>yolov8n-det.imx943.tflite</td></tr>
-            <tr><td>Hailo HEF</td><td>{version}{size}-{task}.hailo{variant}.hef</td><td>yolov8n-det.hailo8l.hef</td></tr>
-            <tr><td>Studio Project</td><td>{Dataset} {Task}</td><td>COCO Detection</td></tr>
-            <tr><td>Studio Experiment</td><td>{Version} {Task}</td><td>YOLOv8 Detection</td></tr>
-        </table>
-
-        <h2>Validation Pipeline</h2>
-        <p>Models go through two validation stages before publication:</p>
-        <table class="arch-table">
-            <tr><th>Stage</th><th>What</th><th>Where</th></tr>
-            <tr>
-                <td>Reference</td>
-                <td>ONNX FP32 and TFLite INT8 mAP on full COCO val2017 (5000 images)</td>
-                <td>EdgeFirst Studio (cloud)</td>
-            </tr>
-            <tr>
-                <td>On-Target</td>
-                <td>Full dataset mAP + timing breakdown (load, preproc, invoke, decode, e2e) per device</td>
-                <td>Board farm (real hardware) <span class="wip-tag">In Progress</span></td>
-            </tr>
-        </table>
-
-        <h2>Perception Architecture</h2>
-        <table class="arch-table">
-            <tr><th>Layer</th><th>Description</th></tr>
-            <tr><td>Foundation</td><td>Hardware abstraction, video I/O, accelerated inference delegates</td></tr>
-            <tr><td>Zenoh</td><td>Modular perception pipeline over Zenoh pub/sub</td></tr>
-            <tr><td>GStreamer</td><td>Spatial perception elements for GStreamer / NNStreamer</td></tr>
-            <tr><td>ROS 2</td><td>Native ROS 2 nodes extending Zenoh microservices <span class="roadmap-tag">Roadmap</span></td></tr>
-        </table>
+        <!-- ═══════════════════════════════════════════════════════ -->
+        <!-- ARCHITECTURE & STUDIO                                  -->
+        <!-- ═══════════════════════════════════════════════════════ -->
 
         <h2>EdgeFirst Studio</h2>
         <p>