Platform docs (#23338)

Signed-off-by: Glenn Jocher <glenn.jocher@ultralytics.com>
Co-authored-by: UltralyticsAssistant <web@ultralytics.com>
This commit is contained in:
Glenn Jocher 2026-01-20 11:51:00 +00:00 committed by GitHub
parent 720d70c2c1
commit d71f7f3266
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
16 changed files with 569 additions and 203 deletions

View File

@ -170,16 +170,22 @@ flowchart LR
Cloud training costs depend on GPU selection:
| GPU | Rate/Hour | Typical Job (1h) |
| --------- | --------- | ---------------- |
| RTX 3090 | $0.92 | $0.92 |
| RTX 4090 | $1.18 | $1.18 |
| L40S | $1.72 | $1.72 |
| RTX 5090 | $1.78 | $1.78 |
| A100 80GB | $2.78 | $2.78 |
| H100 PCIe | $4.78 | $4.78 |
| H100 SXM | $5.38 | $5.38 |
| B200 | $10.38 | $10.38 |
| Tier | GPU | VRAM | Rate/Hour | Typical Job (1h) |
| ---------- | ------------ | ------ | --------- | ---------------- |
| Budget | RTX A2000 | 6 GB | $0.12 | $0.12 |
| Budget | RTX 3080 | 10 GB | $0.25 | $0.25 |
| Budget | RTX 3080 Ti | 12 GB | $0.30 | $0.30 |
| Budget | A30 | 24 GB | $0.44 | $0.44 |
| Mid | L4 | 24 GB | $0.54 | $0.54 |
| Mid | RTX 4090 | 24 GB | $0.60 | $0.60 |
| Mid | A6000 | 48 GB | $0.90 | $0.90 |
| Mid | L40S | 48 GB | $1.72 | $1.72 |
| Pro | A100 40GB | 40 GB | $2.78 | $2.78 |
| Pro | A100 80GB | 80 GB | $3.44 | $3.44 |
| Pro | RTX PRO 6000 | 48 GB | $3.68 | $3.68 |
| Pro | H100 | 80 GB | $5.38 | $5.38 |
| Enterprise | H200 | 141 GB | $5.38 | $5.38 |
| Enterprise | B200 | 192 GB | $10.38 | $10.38 |
See [Cloud Training](../train/cloud-training.md) for complete GPU options and pricing.

View File

@ -61,11 +61,11 @@ View your data region:
<!-- Screenshot: platform-settings-region.avif -->
| Region | Location |
| ------ | ------------------------ |
| **US** | US Central (Iowa) |
| **EU** | EU West (Belgium) |
| **AP** | Asia Pacific (Hong Kong) |
| Region | Location | Best For |
| ------ | -------------------- | --------------------------------------- |
| **US** | Iowa, USA | Americas users, fastest for Americas |
| **EU** | Belgium, Europe | European users, GDPR compliance |
| **AP** | Taiwan, Asia-Pacific | Asia-Pacific users, lowest APAC latency |
!!! note "Region is Permanent"

View File

@ -683,29 +683,148 @@ DELETE /api/api-keys/{keyId}
| `RATE_LIMITED` | Too many requests |
| `INTERNAL_ERROR` | Server error |
## SDK Support
## Python Integration
For easier integration, use the Ultralytics Python package.
### Installation & Setup
```bash
pip install ultralytics
```
Verify installation:
```bash
yolo check
```
!!! warning "Package Version Requirement"
Platform integration requires **ultralytics>=8.4.0**. Lower versions will NOT work with Platform.
```bash
pip install "ultralytics>=8.4.0"
```
### Authentication
**Method 1: CLI Configuration (Recommended)**
```bash
yolo settings api_key=YOUR_API_KEY
```
**Method 2: Environment Variable**
```bash
export ULTRALYTICS_API_KEY=YOUR_API_KEY
```
**Method 3: In Code**
```python
import os
from ultralytics import settings
settings.api_key = "YOUR_API_KEY"
```
### Using Platform Datasets
Reference datasets with `ul://` URIs:
```python
from ultralytics import YOLO
# Set API key
os.environ["ULTRALYTICS_API_KEY"] = "ul_your_key"
model = YOLO("yolo11n.pt")
# Train with Platform integration
model = YOLO("yolo26n.pt")
model.train(data="ul://username/datasets/my-dataset", project="username/my-project", name="experiment-1", epochs=100)
# Train on your Platform dataset
model.train(
data="ul://your-username/your-dataset",
epochs=100,
imgsz=640,
)
```
**URI Format:**
```
ul://{username}/{resource-type}/{name}
Examples:
ul://john/datasets/coco-custom # Dataset
ul://john/my-project # Project
ul://john/my-project/exp-1 # Specific model
ul://ultralytics/yolo26/yolo26n # Official model
```
### Pushing to Platform
Send results to a Platform project:
```python
from ultralytics import YOLO
model = YOLO("yolo11n.pt")
# Results automatically sync to Platform
model.train(
data="coco8.yaml",
epochs=100,
project="ul://your-username/my-project",
name="experiment-1",
)
```
**What syncs:**
- Training metrics (real-time)
- Final model weights
- Validation plots
- Console output
- System metrics
### API Examples
**Load a model from Platform:**
```python
# Your own model
model = YOLO("ul://username/project/model-name")
# Official model
model = YOLO("ul://ultralytics/yolo26/yolo26n")
```
**Run inference:**
```python
results = model("image.jpg")
# Access results
for r in results:
boxes = r.boxes # Detection boxes
masks = r.masks # Segmentation masks
keypoints = r.keypoints # Pose keypoints
probs = r.probs # Classification probabilities
```
**Export model:**
```python
# Export to ONNX
model.export(format="onnx", imgsz=640, half=True)
# Export to TensorRT
model.export(format="engine", imgsz=640, half=True)
# Export to CoreML
model.export(format="coreml", imgsz=640)
```
**Validation:**
```python
metrics = model.val(data="ul://username/my-dataset")
print(f"mAP50: {metrics.box.map50}")
print(f"mAP50-95: {metrics.box.map}")
```
## Webhooks

View File

@ -34,6 +34,58 @@ The annotation editor supports all 5 YOLO task types:
| **OBB** | Oriented Box | Rotated bounding boxes (4 corners) |
| **Classify** | Class Selector | Image-level labels |
### Task Details
??? info "Object Detection"
**What it does:** Identifies objects and their locations with axis-aligned bounding boxes.
**Label format:** `class_id center_x center_y width height` (all normalized 0-1)
**Example:** `0 0.5 0.5 0.2 0.3` — Class 0 centered at (50%, 50%) with 20% width and 30% height
**Use cases:** Inventory counting, traffic monitoring, wildlife detection, security systems
??? info "Instance Segmentation"
**What it does:** Creates pixel-precise masks for each object instance.
**Label format:** `class_id x1 y1 x2 y2 x3 y3 ...` (polygon vertices, normalized 0-1)
**Example:** `0 0.1 0.1 0.9 0.1 0.9 0.9 0.1 0.9` — Quadrilateral mask
**Use cases:** Medical imaging, autonomous vehicles, photo editing, agricultural analysis
??? info "Pose Estimation"
**What it does:** Detects body keypoints for skeleton tracking.
**Label format:** `class_id cx cy w h kx1 ky1 v1 kx2 ky2 v2 ...`
- Visibility flags: `0`=not labeled, `1`=labeled but occluded, `2`=labeled and visible
**Example:** `0 0.5 0.5 0.2 0.3 0.6 0.7 2 0.4 0.8 1` — Person with 2 keypoints
**Use cases:** Sports analysis, physical therapy, animation, gesture recognition
??? info "Oriented Bounding Box (OBB)"
**What it does:** Detects rotated objects with angle-aware bounding boxes.
**Label format:** `class_id x1 y1 x2 y2 x3 y3 x4 y4` (four corner points, normalized)
**Example:** `0 0.1 0.1 0.9 0.1 0.9 0.9 0.1 0.9` — Rotated rectangle
**Use cases:** Aerial imagery, document analysis, manufacturing inspection, ship detection
??? info "Image Classification"
**What it does:** Assigns a single label to the entire image.
**Label format:** Folder-based — images organized by class name (`train/cats/`, `train/dogs/`)
**Use cases:** Content moderation, quality control, medical diagnosis, scene recognition
## Getting Started
To annotate images:

View File

@ -8,37 +8,49 @@ keywords: Ultralytics Platform, datasets, dataset management, YOLO, data upload,
[Ultralytics Platform](https://platform.ultralytics.com) datasets provide a streamlined solution for managing your training data. Once uploaded, datasets can be immediately used for model training, with automatic processing and statistics generation.
<p align="center">
<iframe loading="lazy" width="720" height="405" src="https://www.youtube.com/embed/R42s2zFtNIY"
title="YouTube video player" frameborder="0"
allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share"
allowfullscreen>
</iframe>
<br>
<strong>Watch:</strong> Upload Datasets to Ultralytics Platform
</p>
## Upload Dataset
Ultralytics Platform accepts multiple upload formats for flexibility:
Ultralytics Platform accepts multiple upload formats for flexibility.
| Format | Description |
| --------------- | -------------------------------------------------- |
| **Images** | Individual image files (JPG, PNG, WebP, TIFF, RAW) |
| **ZIP Archive** | Compressed folder with images and optional labels |
| **Video** | MP4, AVI files - frames extracted at ~1 fps |
| **YOLO Format** | Standard YOLO directory structure with labels |
### Supported Image Formats
### Video Frame Extraction
| Format | Extensions | Notes |
| ------ | --------------- | ------------------------ |
| JPEG | `.jpg`, `.jpeg` | Most common, recommended |
| PNG | `.png` | Supports transparency |
| WebP | `.webp` | Modern, good compression |
| BMP | `.bmp` | Uncompressed |
| GIF | `.gif` | First frame extracted |
| TIFF | `.tiff`, `.tif` | High quality |
| HEIC | `.heic` | iPhone photos |
| AVIF | `.avif` | Next-gen format |
| PFM | `.pfm` | Floating point |
| DNG | `.dng` | Raw camera |
When uploading videos, frames are automatically extracted:
### Supported Video Formats
- **Frame rate**: ~1 frame per second
- **Maximum frames**: 100 frames per video
- **Processing**: Client-side extraction before upload
- **Format**: Frames converted to standard image format
Videos are automatically extracted to frames:
This is ideal for creating training datasets from surveillance footage, action recordings, or any video source.
| Format | Extensions | Extraction |
| ------ | ---------- | --------------------- |
| MP4 | `.mp4` | 1 FPS, max 100 frames |
| WebM | `.webm` | 1 FPS, max 100 frames |
| MOV | `.mov` | 1 FPS, max 100 frames |
| AVI | `.avi` | 1 FPS, max 100 frames |
| MKV | `.mkv` | 1 FPS, max 100 frames |
| M4V | `.m4v` | 1 FPS, max 100 frames |
### File Size Limits
| Type | Maximum Size |
| --------- | ------------ |
| Images | 50 MB each |
| Videos | 1 GB each |
| ZIP files | 50 GB |
### Archives
ZIP files up to 50GB are supported with folder structure preserved and automatic extraction and processing.
### Preparing Your Dataset
@ -294,9 +306,12 @@ Use the bulk selection feature:
Ultralytics Platform supports YOLO format labels:
- **Detect**: `class_id x_center y_center width height`
- **Segment**: `class_id x1 y1 x2 y2 ...` (polygon points)
- **Pose**: `class_id x_center y_center width height kp1_x kp1_y kp1_v ...`
- **OBB**: `class_id x1 y1 x2 y2 x3 y3 x4 y4`
| Task | Format | Example |
| -------- | -------------------------------- | ----------------------------------- |
| Detect | `class cx cy w h` | `0 0.5 0.5 0.2 0.3` |
| Segment | `class x1 y1 x2 y2 ...` | `0 0.1 0.1 0.9 0.1 0.9 0.9` |
| Pose | `class cx cy w h kx1 ky1 v1 ...` | `0 0.5 0.5 0.2 0.3 0.6 0.7 2` |
| OBB | `class x1 y1 x2 y2 x3 y3 x4 y4` | `0 0.1 0.1 0.9 0.1 0.9 0.9 0.1 0.9` |
| Classify | Directory structure | `train/cats/`, `train/dogs/` |
All coordinates are normalized (0-1 range).
All coordinates are normalized (0-1 range). Pose visibility flags: 0=not labeled, 1=labeled but occluded, 2=labeled and visible.

View File

@ -92,9 +92,11 @@ Every dataset includes automatic statistics:
Ultralytics Platform supports:
- **Images**: JPG, PNG, WebP, TIFF, BMP, and other common formats
- **Videos**: MP4, AVI, MOV - frames are extracted automatically
- **Archives**: ZIP files containing images with optional YOLO-format labels
**Images:** JPEG, PNG, WebP, BMP, GIF, TIFF, HEIC, AVIF, PFM, DNG (max 50MB each)
**Videos:** MP4, WebM, MOV, AVI, MKV, M4V (max 1GB, frames extracted at 1 FPS, max 100 frames)
**Archives:** ZIP files (max 50GB) containing images with optional YOLO-format labels
### What is the maximum dataset size?
@ -106,6 +108,8 @@ Storage limits depend on your plan:
| Pro | 500 GB |
| Enterprise | Custom |
Individual file limits: Images 50MB, Videos 1GB, ZIP archives 50GB
### Can I use my Platform datasets for local training?
Yes! Use the dataset URI format to train locally:

View File

@ -52,19 +52,19 @@ View all regions with details:
## Available Regions
### Americas (15 regions)
### Americas (14 regions)
| Region | Location |
| Zone | Location |
| ----------------------- | ------------------- |
| us-central1 | Iowa, USA |
| us-east1 | South Carolina, USA |
| us-east4 | Virginia, USA |
| us-east5 | Columbus, USA |
| us-south1 | Dallas, USA |
| us-east5 | Ohio, USA |
| us-west1 | Oregon, USA |
| us-west2 | Los Angeles, USA |
| us-west3 | Salt Lake City, USA |
| us-west4 | Las Vegas, USA |
| us-south1 | Dallas, USA |
| northamerica-northeast1 | Montreal, Canada |
| northamerica-northeast2 | Toronto, Canada |
| southamerica-east1 | São Paulo, Brazil |
@ -72,11 +72,8 @@ View all regions with details:
### Europe (12 regions)
| Region | Location |
| Zone | Location |
| ----------------- | ------------------- |
| europe-central2 | Warsaw, Poland |
| europe-north1 | Finland |
| europe-southwest1 | Madrid, Spain |
| europe-west1 | Belgium |
| europe-west2 | London, UK |
| europe-west3 | Frankfurt, Germany |
@ -86,25 +83,33 @@ View all regions with details:
| europe-west9 | Paris, France |
| europe-west10 | Berlin, Germany |
| europe-west12 | Turin, Italy |
| europe-north1 | Finland |
| europe-central2 | Warsaw, Poland |
| europe-southwest1 | Madrid, Spain |
### Asia Pacific (16 regions)
### Asia-Pacific (14 regions)
| Region | Location |
| Zone | Location |
| -------------------- | -------------------- |
| asia-east1 | Taiwan |
| asia-east2 | Hong Kong |
| asia-northeast1 | Tokyo, Japan |
| asia-northeast2 | Osaka, Japan |
| asia-northeast3 | Seoul, Korea |
| asia-northeast3 | Seoul, South Korea |
| asia-south1 | Mumbai, India |
| asia-south2 | Delhi, India |
| asia-southeast1 | Singapore |
| asia-southeast2 | Jakarta, Indonesia |
| australia-southeast1 | Sydney, Australia |
| australia-southeast2 | Melbourne, Australia |
| me-central1 | Doha, Qatar |
| me-central2 | Dammam, Saudi Arabia |
| me-west1 | Tel Aviv, Israel |
### Middle East & Africa (3 regions)
| Zone | Location |
| ----------- | -------------------- |
| me-central1 | Doha, Qatar |
| me-central2 | Dammam, Saudi Arabia |
| me-west1 | Tel Aviv, Israel |
## Endpoint Configuration

View File

@ -56,11 +56,11 @@ graph LR
The shared inference service runs in 3 key regions:
| Region | Location |
| ------ | ------------------------ |
| US | US Central (Iowa) |
| EU | EU West (Belgium) |
| AP | Asia Pacific (Hong Kong) |
| Region | Location |
| ------ | -------------------- |
| US | Iowa, USA |
| EU | Belgium, Europe |
| AP | Taiwan, Asia-Pacific |
Requests are routed to your data region automatically.

View File

@ -57,7 +57,7 @@ Adjust detection behavior with parameters:
| Parameter | Range | Default | Description |
| -------------- | ------- | ------- | ---------------------------- |
| **Confidence** | 0.0-1.0 | 0.25 | Minimum confidence threshold |
| **IoU** | 0.0-1.0 | 0.45 | NMS IoU threshold |
| **IoU** | 0.0-1.0 | 0.70 | NMS IoU threshold |
| **Image Size** | 32-1280 | 640 | Input resize dimension |
### Confidence Threshold
@ -72,9 +72,9 @@ Filter predictions by confidence:
Control Non-Maximum Suppression:
- **Higher (0.7+)**: Allow overlapping boxes
- **Lower (0.3-0.45)**: Merge nearby detections
- **Default (0.45)**: Standard NMS behavior
- **Higher (0.7+)**: Allow more overlapping boxes
- **Lower (0.3-0.5)**: Merge nearby detections more aggressively
- **Default (0.70)**: Balanced NMS behavior for most use cases
## REST API
@ -104,7 +104,7 @@ POST https://platform.ultralytics.com/api/models/{model_slug}/predict
-H "Authorization: Bearer YOUR_API_KEY" \
-F "file=@image.jpg" \
-F "conf=0.25" \
-F "iou=0.45"
-F "iou=0.7"
```
=== "Python"
@ -115,7 +115,7 @@ POST https://platform.ultralytics.com/api/models/{model_slug}/predict
url = "https://platform.ultralytics.com/api/models/username/project/model/predict"
headers = {"Authorization": "Bearer YOUR_API_KEY"}
files = {"file": open("image.jpg", "rb")}
data = {"conf": 0.25, "iou": 0.45}
data = {"conf": 0.25, "iou": 0.7}
response = requests.post(url, headers=headers, files=files, data=data)
print(response.json())

View File

@ -108,6 +108,35 @@ Copy a public project:
2. Click **Fork**
3. Project copies with all models
## Official Ultralytics Models
Featured at the top of Explore, you'll find official Ultralytics models:
| Project | Description | Models |
| ---------- | --------------------------- | ---------------------------- |
| **YOLO26** | Latest January 2026 release | 27 models (all sizes, tasks) |
| **YOLO11** | Current stable release | 10+ models |
| **YOLOv8** | Previous generation | Various |
| **YOLOv5** | Legacy, widely adopted | Various |
**Project Cards Show:**
- Project icon and name
- Public badge
- Creator avatar and username
- Short description
- Model count and total size
- Last updated
- Model name tags
**Dataset Cards Show:**
- Dataset name
- Task type badge
- Creator info
- Image count
- Preview thumbnails
## User Profiles
View public profiles:

View File

@ -7,7 +7,6 @@ keywords: Ultralytics Platform, YOLO, computer vision, model training, cloud dep
# Ultralytics Platform
<div align="center">
<!-- Screenshot: platform-banner.avif -->
<a href="https://docs.ultralytics.com/zh/platform/">中文</a> |
<a href="https://docs.ultralytics.com/ko/platform/">한국어</a> |
<a href="https://docs.ultralytics.com/ja/platform/">日本語</a> |
@ -28,15 +27,7 @@ keywords: Ultralytics Platform, YOLO, computer vision, model training, cloud dep
[Ultralytics Platform](https://platform.ultralytics.com) is a comprehensive end-to-end computer vision platform that streamlines the entire ML workflow from data preparation to model deployment. Built for teams and individuals who need production-ready [computer vision](https://www.ultralytics.com/glossary/computer-vision-cv) solutions without the infrastructure complexity.
<p align="center">
<iframe loading="lazy" width="720" height="405" src="https://www.youtube.com/embed/qE-dfbB5Sis"
title="YouTube video player" frameborder="0"
allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share"
allowfullscreen>
</iframe>
<br>
<strong>Watch:</strong> Getting Started with Ultralytics Platform
</p>
![Ultralytics Platform Dataset Screenshot](https://cdn.jsdelivr.net/gh/ultralytics/assets@main/docs/platform/dataset-screenshot.avif)
## What is Ultralytics Platform?
@ -50,9 +41,9 @@ Ultralytics Platform is designed to replace fragmented ML tooling with a unified
All in one platform with native support for YOLO26 and YOLO11 models.
## Workflow: Data → Train → Deploy
## Workflow: Upload → Annotate → Train → Export → Deploy
The Platform follows a streamlined three-stage workflow:
The Platform provides an end-to-end workflow:
```mermaid
graph LR
@ -65,27 +56,39 @@ graph LR
E --> F[View Metrics]
end
subgraph Deploy["🌐 Deploy"]
G[Test] --> H[Deploy Endpoint]
G[Export] --> H[Deploy Endpoint]
H --> I[Monitor]
end
Data --> Train --> Deploy
```
| Stage | Features |
| ---------- | --------------------------------------------------------------------------------------- |
| **Data** | Dataset upload, annotation editor (manual + SAM + auto-label), statistics visualization |
| **Train** | Project organization, cloud training with GPU instances, real-time metrics streaming |
| **Deploy** | Inference testing, dedicated endpoints in 43 regions, monitoring dashboards |
| Stage | Features |
| ------------ | --------------------------------------------------------------------------- |
| **Upload** | Images (50MB), videos (1GB), ZIP archives (50GB) with automatic processing |
| **Annotate** | Manual tools, SAM smart annotation, YOLO auto-labeling for all 5 task types |
| **Train** | Cloud GPUs (RTX 4090 to H200), real-time metrics, project organization |
| **Export** | 17 deployment formats (ONNX, TensorRT, CoreML, TFLite, etc.) |
| **Deploy** | 43 global regions with dedicated endpoints, auto-scaling, monitoring |
**What you can do:**
- **Upload** images, videos, and ZIP archives to create training datasets
- **Visualize** annotations with interactive overlays for all 5 YOLO task types
- **Train** models on cloud GPUs (RTX 4090 to H200) with real-time metrics
- **Export** to 17 deployment formats (ONNX, TensorRT, CoreML, TFLite, etc.)
- **Deploy** to 43 global regions with one-click dedicated endpoints
- **Monitor** training progress, deployment health, and usage metrics
- **Collaborate** by making projects and datasets public for the community
## Multi-Region Infrastructure
Your data stays in your region. Ultralytics Platform operates infrastructure in three global regions:
| Region | Location |
| ------ | ------------------------ |
| **US** | US Central (Iowa) |
| **EU** | EU West (Belgium) |
| **AP** | Asia Pacific (Hong Kong) |
| Region | Location | Best For |
| ------ | -------------------- | --------------------------------------- |
| **US** | Iowa, USA | Americas users, fastest for Americas |
| **EU** | Belgium, Europe | European users, GDPR compliance |
| **AP** | Taiwan, Asia-Pacific | Asia-Pacific users, lowest APAC latency |
You select your region during onboarding, and all your data, models, and deployments remain in that region.
@ -106,6 +109,8 @@ You select your region during onboarding, and all your data, models, and deploym
- **Project Organization**: Group related models, compare experiments, track activity
- **17 Export Formats**: ONNX, TensorRT, CoreML, TFLite, and more
![Ultralytics Platform Project Screenshot](https://cdn.jsdelivr.net/gh/ultralytics/assets@main/docs/platform/project-screenshot.avif)
### Deployment
- **Inference Testing**: Test models directly in the browser with custom images
@ -168,13 +173,22 @@ For a detailed guide, see the [Quickstart](quickstart.md) page.
Ultralytics Platform supports multiple GPU types for cloud training:
| GPU | VRAM | Cost | Best For |
| --------- | ----- | --------- | ------------------------- |
| RTX 4090 | 24GB | $1.18/hr | Fast training, best value |
| L40S | 48GB | $1.72/hr | Balanced performance |
| A100 80GB | 80GB | $2.78/hr | Large models, big batches |
| H100 PCIe | 80GB | $4.78/hr | Maximum performance |
| B200 | 180GB | $10.38/hr | Cutting-edge workloads |
| Tier | GPU | VRAM | Cost/Hour | Best For |
| ----------- | ------------ | ------ | --------- | -------------------------- |
| Budget | RTX A2000 | 6 GB | $0.12 | Small datasets, testing |
| Budget | RTX 3080 | 10 GB | $0.25 | Medium datasets |
| Budget | RTX 3080 Ti | 12 GB | $0.30 | Medium datasets |
| Budget | A30 | 24 GB | $0.44 | Larger batch sizes |
| Mid | RTX 4090 | 24 GB | $0.60 | Great price/performance |
| Mid | A6000 | 48 GB | $0.90 | Large models |
| Mid | L4 | 24 GB | $0.54 | Inference optimized |
| Mid | L40S | 48 GB | $1.72 | Large batch training |
| Pro | A100 40GB | 40 GB | $2.78 | Production training |
| Pro | A100 80GB | 80 GB | $3.44 | Very large models |
| Pro | H100 | 80 GB | $5.38 | Fastest training |
| Enterprise | H200 | 141 GB | $5.38 | Maximum performance |
| Enterprise | B200 | 192 GB | $10.38 | Largest models |
| Ultralytics | RTX PRO 6000 | 48 GB | $3.68 | Ultralytics infrastructure |
See [Cloud Training](train/cloud-training.md) for complete pricing and GPU options.
@ -210,3 +224,70 @@ The Platform includes a full-featured annotation editor supporting:
- **Keyboard Shortcuts**: Efficient workflows with hotkeys
See [Annotation](data/annotation.md) for the complete guide.
## Troubleshooting
### Dataset Issues
| Problem | Solution |
| ---------------------- | -------------------------------------------------------------------------------------------------------- |
| Dataset won't process | Check file format is supported (JPEG, PNG, WebP, etc.). Max file size: images 50MB, videos 1GB, ZIP 50GB |
| Missing annotations | Verify labels are in YOLO format with `.txt` files matching image filenames |
| "Train split required" | Add `train/` folder to your dataset structure, or create splits in the dataset settings |
| Class names undefined | Add a `data.yaml` file with `names:` list, or define classes in dataset settings |
### Training Issues
| Problem | Solution |
| -------------------- | ----------------------------------------------------------------------------------- |
| Training won't start | Check credit balance in Settings > Billing. Minimum $5.00 required |
| Out of memory error | Reduce batch size, use smaller model (n/s), or select GPU with more VRAM |
| Poor metrics | Check dataset quality, increase epochs, try data augmentation, verify class balance |
| Training slow | Select faster GPU, reduce image size, check dataset isn't bottlenecked |
### Deployment Issues
| Problem | Solution |
| ----------------------- | -------------------------------------------------------------------------------------- |
| Endpoint not responding | Check endpoint status (Running vs Stopped). Cold start may take 2-5 seconds |
| 401 Unauthorized | Verify API key is correct and has required scopes |
| Slow inference | Check model size, consider TensorRT export, select closer region |
| Export failed | Some formats require specific model architectures. Try ONNX for broadest compatibility |
### Common Questions
??? question "Can I change my username after signup?"
No, usernames are permanent and cannot be changed. Choose carefully during signup.
??? question "Can I change my data region?"
No, data region is selected during signup and cannot be changed. To switch regions, create a new account and re-upload your data.
??? question "How do I get more credits?"
Go to Settings > Billing > Add Credits. Purchase credits from $5 to $1000. Purchased credits never expire.
??? question "What happens if training fails?"
You're only charged for completed compute time. Checkpoints are saved, and you can resume training.
??? question "Can I download my trained model?"
Yes, click the download icon on any model page to download the `.pt` file or exported formats.
??? question "How do I share my work publicly?"
Edit your project or dataset settings and toggle visibility to "Public". Public content appears on the Explore page.
??? question "What are the file size limits?"
Images: 50MB, Videos: 1GB, ZIP archives: 50GB. For larger files, split into multiple uploads.
??? question "How long are deleted items kept in Trash?"
30 days. After that, items are permanently deleted and cannot be recovered.
??? question "Can I use Platform models commercially?"
Free and Pro plans use AGPL license. For commercial use without AGPL requirements, contact sales@ultralytics.com for Enterprise licensing.

View File

@ -25,16 +25,6 @@ journey
Deploy endpoint: 5: User
```
<p align="center">
<iframe loading="lazy" width="720" height="405" src="https://www.youtube.com/embed/qE-dfbB5Sis"
title="YouTube video player" frameborder="0"
allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share"
allowfullscreen>
</iframe>
<br>
<strong>Watch:</strong> Getting Started with Ultralytics Platform
</p>
## Get Started
[Ultralytics Platform](https://platform.ultralytics.com) offers a variety of easy signup options. You can register and log in using your Google, Apple, or GitHub accounts, or simply with your email address.
@ -47,16 +37,29 @@ During signup, you'll be asked to select your data region. This is an important
<!-- Screenshot: platform-onboarding-region.avif -->
| Region | Location | Best For |
| ------ | ------------------------ | ------------------------------- |
| **US** | US Central (Iowa) | Americas users |
| **EU** | EU West (Belgium) | European users, GDPR compliance |
| **AP** | Asia Pacific (Hong Kong) | Asia-Pacific users |
| Region | Location | Best For |
| ------ | -------------------- | --------------------------------------- |
| **US** | Iowa, USA | Americas users, fastest for Americas |
| **EU** | Belgium, Europe | European users, GDPR compliance |
| **AP** | Taiwan, Asia-Pacific | Asia-Pacific users, lowest APAC latency |
!!! warning "Region is Permanent"
Your region selection cannot be changed after account creation. Choose the region closest to you or your users for best performance.
### Free Credits
Every new account receives free credits for cloud GPU training:
| Email Type | Sign-up Credits | How to Qualify |
| ---------------------- | --------------- | -------------------------------------- |
| **Work/Company Email** | **$25.00** | Use your company domain (@company.com) |
| **Personal Email** | **$5.00** | Gmail, Yahoo, Outlook, etc. |
!!! tip "Maximize Your Credits"
Sign up with a work email to receive $25 in credits. If you signed up with a personal email, you can verify a work email later to unlock the additional $20 in credits.
### Complete Your Profile
After selecting your region, complete your profile with your information.
@ -73,13 +76,31 @@ After signing in, you will be directed to the Home page of [Ultralytics Platform
<!-- Screenshot: platform-dashboard.avif -->
The sidebar provides access to key sections:
The sidebar provides access to all Platform sections:
- **Home**: Dashboard with quick actions and recent items
- **Datasets**: Your uploaded datasets
- **Projects**: Your training projects and models
- **Deployments**: Your model endpoints
- **Explore**: Public content from the community
**Top Section:**
| Item | Description |
| ----------- | ------------------------------------------------ |
| **Search** | Quick search across all your resources (Cmd+K) |
| **Home** | Dashboard with quick actions and recent activity |
| **Explore** | Discover public projects and datasets |
**My Workspace:**
| Section | Description |
| ------------ | --------------------------------------- |
| **Annotate** | Your datasets organized for annotation |
| **Train** | Your projects containing trained models |
| **Deploy** | Your active deployments |
**Bottom Section:**
| Item | Description |
| ------------ | --------------------------------------- |
| **Trash** | Deleted items (recoverable for 30 days) |
| **Settings** | Account, billing, and preferences |
| **Feedback** | Send feedback to Ultralytics |
### Quick Actions

View File

@ -8,16 +8,6 @@ keywords: Ultralytics Platform, cloud training, GPU training, remote training, Y
[Ultralytics Platform](https://platform.ultralytics.com) Cloud Training offers single-click training on cloud GPUs, making model training accessible without complex setup. Train YOLO models with real-time metrics streaming and automatic checkpoint saving.
<p align="center">
<iframe loading="lazy" width="720" height="405" src="https://www.youtube.com/embed/ie3vLUDNYZo"
title="YouTube video player" frameborder="0"
allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share"
allowfullscreen>
</iframe>
<br>
<strong>Watch:</strong> Cloud Training with Ultralytics Platform
</p>
## Train from UI
Start cloud training directly from the Platform:
@ -57,25 +47,28 @@ Choose your compute resources:
<!-- Screenshot: platform-training-gpu.avif -->
| GPU | VRAM | Speed | Cost/Hour |
| -------------- | ----- | --------- | --------- |
| RTX 2000 Ada | 16GB | Good | $0.48 |
| RTX A4000 | 16GB | Good | $0.50 |
| RTX 3090 | 24GB | Good | $0.92 |
| RTX A6000 | 48GB | Fast | $0.98 |
| RTX 4090 | 24GB | Fast | $1.18 |
| L40S | 48GB | Fast | $1.72 |
| RTX 5090 | 32GB | Very Fast | $1.78 |
| A100 80GB PCIe | 80GB | Very Fast | $2.78 |
| H100 PCIe | 80GB | Fastest | $4.78 |
| H100 SXM | 80GB | Fastest | $5.38 |
| B200 | 180GB | Fastest | $10.38 |
| Tier | GPU | VRAM | Price/Hour | Best For |
| ----------- | ------------ | ------ | ---------- | -------------------------- |
| Budget | RTX A2000 | 6 GB | $0.12 | Small datasets, testing |
| Budget | RTX 3080 | 10 GB | $0.25 | Medium datasets |
| Budget | RTX 3080 Ti | 12 GB | $0.30 | Medium datasets |
| Budget | A30 | 24 GB | $0.44 | Larger batch sizes |
| Mid | RTX 4090 | 24 GB | $0.60 | Great price/performance |
| Mid | A6000 | 48 GB | $0.90 | Large models |
| Mid | L4 | 24 GB | $0.54 | Inference optimized |
| Mid | L40S | 48 GB | $1.72 | Large batch training |
| Pro | A100 40GB | 40 GB | $2.78 | Production training |
| Pro | A100 80GB | 80 GB | $3.44 | Very large models |
| Pro | H100 | 80 GB | $5.38 | Fastest training |
| Enterprise | H200 | 141 GB | $5.38 | Maximum performance |
| Enterprise | B200 | 192 GB | $10.38 | Largest models |
| Ultralytics | RTX PRO 6000 | 48 GB | $3.68 | Ultralytics infrastructure |
!!! tip "GPU Selection"
- **RTX 4090**: Best value with excellent performance for most jobs
- **RTX 4090**: Best price/performance ratio for most jobs at $0.60/hr
- **A100 80GB**: Required for large batch sizes or big models
- **H100**: Maximum performance for time-sensitive training
- **H100/H200**: Maximum performance for time-sensitive training
- **B200**: NVIDIA Blackwell architecture for cutting-edge workloads
### Step 4: Start Training
@ -201,17 +194,45 @@ The `ul://` URI format automatically downloads and configures your dataset.
Training costs are based on GPU usage:
### Cost Calculation
### Cost Estimation
Before training starts, the Platform estimates total cost based on:
```
Total Cost = GPU Rate × Training Time (hours)
Estimated Cost = Base Time × Model Multiplier × Dataset Multiplier × GPU Speed Factor × GPU Rate
```
| Example | GPU | Time | Cost |
| ---------- | --------- | ------- | ------ |
| Small job | RTX 4090 | 1 hour | $1.18 |
| Medium job | A100 80GB | 4 hours | $11.12 |
| Large job | H100 PCIe | 8 hours | $38.24 |
**Factors affecting cost:**
| Factor | Impact |
| -------------------- | ------------------------------------------------ |
| **Dataset Size** | More images = longer training time |
| **Model Size** | Larger models (m, l, x) train slower than (n, s) |
| **Number of Epochs** | Direct multiplier on training time |
| **Image Size** | Larger imgsz increases computation |
| **GPU Speed** | Faster GPUs reduce training time |
### Cost Examples
| Scenario | GPU | Time | Cost |
| --------------------------------- | --------- | -------- | ------- |
| 1000 images, YOLO26n, 100 epochs | RTX 4090 | ~1 hour | ~$0.60 |
| 5000 images, YOLO26m, 100 epochs | A100 80GB | ~4 hours | ~$13.76 |
| 10000 images, YOLO26x, 200 epochs | H100 | ~8 hours | ~$43.04 |
### Hold/Settle System
The Platform uses a consumer-protection billing model:
1. **Estimate**: Cost calculated before training starts
2. **Hold**: Estimated amount + 20% safety margin reserved from balance
3. **Train**: Reserved amount shown as "Reserved" in your balance
4. **Settle**: After completion, charged only for actual GPU time used
5. **Refund**: Any excess automatically returned to your balance
!!! success "Consumer Protection"
You're **never charged more than the estimate** shown before training. If training completes early or is canceled, you only pay for actual compute time used.
### Payment Methods

View File

@ -57,14 +57,22 @@ Ultralytics Platform supports multiple training approaches:
Available GPUs for cloud training:
| GPU | VRAM | Performance | Cost |
| --------- | ----- | ----------- | --------- |
| RTX 3090 | 24GB | Good | $0.92/hr |
| RTX 4090 | 24GB | Excellent | $1.18/hr |
| L40S | 48GB | Very Good | $1.72/hr |
| A100 80GB | 80GB | Excellent | $2.78/hr |
| H100 PCIe | 80GB | Best | $4.78/hr |
| B200 | 180GB | Best | $10.38/hr |
| Tier | GPU | VRAM | Cost/Hour | Best For |
| ---------- | ------------ | ------ | --------- | -------------------------- |
| Budget | RTX A2000 | 6 GB | $0.12 | Small datasets, testing |
| Budget | RTX 3080 | 10 GB | $0.25 | Medium datasets |
| Budget | RTX 3080 Ti | 12 GB | $0.30 | Medium datasets |
| Budget | A30 | 24 GB | $0.44 | Larger batch sizes |
| Mid | L4 | 24 GB | $0.54 | Inference optimized |
| Mid | RTX 4090 | 24 GB | $0.60 | Great price/performance |
| Mid | A6000 | 48 GB | $0.90 | Large models |
| Mid | L40S | 48 GB | $1.72 | Large batch training |
| Pro | A100 40GB | 40 GB | $2.78 | Production training |
| Pro | A100 80GB | 80 GB | $3.44 | Very large models |
| Pro | RTX PRO 6000 | 48 GB | $3.68 | Ultralytics infrastructure |
| Pro | H100 | 80 GB | $5.38 | Fastest training |
| Enterprise | H200 | 141 GB | $5.38 | Maximum performance |
| Enterprise | B200 | 192 GB | $10.38 | Largest models |
!!! tip "Signup Credits"

View File

@ -120,27 +120,42 @@ Export your model to 17 deployment formats:
<!-- Screenshot: platform-models-export.avif -->
### Supported Formats
### Supported Formats (17 total)
| Format | Description | Use Case |
| ----------------- | ---------------------------- | ------------------------- |
| **ONNX** | Open Neural Network Exchange | Cross-platform deployment |
| **TorchScript** | Serialized PyTorch | PyTorch deployment |
| **OpenVINO** | Intel optimization | Intel CPUs/GPUs |
| **TensorRT** | NVIDIA optimization | NVIDIA GPUs |
| **CoreML** | Apple optimization | iOS/macOS |
| **TFLite** | TensorFlow Lite | Mobile/embedded |
| **TF SavedModel** | TensorFlow format | TensorFlow ecosystem |
| **TF GraphDef** | TensorFlow frozen | Legacy TensorFlow |
| **PaddlePaddle** | Baidu framework | PaddlePaddle ecosystem |
| **NCNN** | Mobile inference | Android/embedded |
| **Edge TPU** | Google Edge TPU | Coral devices |
| **TF.js** | TensorFlow.js | Browser deployment |
| **MNN** | Alibaba framework | Mobile optimization |
| **RKNN** | Rockchip NPU | Rockchip devices |
| **IMX** | NXP i.MX | NXP platforms |
| **Axelera** | Metis AI | Edge AI accelerators |
| **ExecuTorch** | Meta framework | Meta platforms |
| # | Format | File Extension | Use Case |
| --- | ----------------- | ---------------- | ---------------------------------- |
| 1 | **ONNX** | `.onnx` | Cross-platform, web, most runtimes |
| 2 | **TorchScript** | `.torchscript` | PyTorch deployment without Python |
| 3 | **OpenVINO** | `.xml`, `.bin` | Intel CPUs, GPUs, VPUs |
| 4 | **TensorRT** | `.engine` | NVIDIA GPUs (fastest inference) |
| 5 | **CoreML** | `.mlpackage` | Apple iOS, macOS, watchOS |
| 6 | **TF Lite** | `.tflite` | Mobile (Android, iOS), edge |
| 7 | **TF SavedModel** | `saved_model/` | TensorFlow Serving |
| 8 | **TF GraphDef** | `.pb` | TensorFlow 1.x |
| 9 | **TF Edge TPU** | `.tflite` | Google Coral devices |
| 10 | **TF.js** | `.json`, `.bin` | Browser inference |
| 11 | **PaddlePaddle** | `.pdmodel` | Baidu PaddlePaddle |
| 12 | **NCNN** | `.param`, `.bin` | Mobile (Android/iOS), optimized |
| 13 | **MNN** | `.mnn` | Alibaba mobile runtime |
| 14 | **RKNN** | `.rknn` | Rockchip NPUs |
| 15 | **IMX500** | `.imx` | Sony IMX500 sensor |
| 16 | **Axelera** | `.axelera` | Axelera AI accelerators |
### Format Selection Guide
**For NVIDIA GPUs:** Use **TensorRT** for maximum speed
**For Intel Hardware:** Use **OpenVINO** for Intel CPUs, GPUs, and VPUs
**For Apple Devices:** Use **CoreML** for iOS, macOS, Apple Silicon
**For Android:** Use **TF Lite** or **NCNN** for best performance
**For Web Browsers:** Use **TF.js** or **ONNX** (with ONNX Runtime Web)
**For Edge Devices:** Use **TF Edge TPU** for Coral, **RKNN** for Rockchip
**For General Compatibility:** Use **ONNX** — works with most inference runtimes
<!-- Screenshot: platform-models-export-progress.avif -->

View File

@ -8,16 +8,6 @@ keywords: Ultralytics Platform, projects, model management, experiment tracking,
[Ultralytics Platform](https://platform.ultralytics.com) projects provide an effective solution for organizing and managing your models. Group related models together to facilitate easier management, comparison, and development.
<p align="center">
<iframe loading="lazy" width="720" height="405" src="https://www.youtube.com/embed/Gc6K5eKrTNQ"
title="YouTube video player" frameborder="0"
allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share"
allowfullscreen>
</iframe>
<br>
<strong>Watch:</strong> Train YOLO Models Using Ultralytics Platform
</p>
## Create Project
Navigate to **Projects** in the sidebar and click **Create Project**.