Everything
Your Data Needs.
Import, curate, augment, and export datasets for any computer vision task — all from a single platform connected to your AI tools.
Every Vision
Task Covered
Native support with type-specific viewers, statistics, and format-aware exports.
Object Detection
Bounding boxes with YOLO, COCO, Pascal VOC formats
Classification
Folder-based, CSV, and HuggingFace structures
Segmentation
Instance and semantic masks in COCO and YOLO Seg
Keypoints / Pose
Skeleton and landmark annotation in COCO and YOLO Pose
OCR
Text detection and recognition with COCO OCR, ICDAR
Video Tracking
Multi-object tracking with MOT and COCO Video
Vision-Language
Image captioning, VQA, and multimodal pairs
3D Point Cloud
PCD, PLY, KITTI formats with spatial operations
Image-to-Image
Paired and unpaired image translation datasets
Video Classification
Folder and CSV-based video categorization
Tabular Data
CSV, XLSX, Parquet with column operations
Powerful
Operations
Every operation runs as an async job with real-time progress and creates an immutable version.
MCP Agent
Integration
Connect cvPal to Claude, Cursor, Gemini, or any MCP-compatible AI. Describe what you need in natural language and the agent runs the tools for you.
- check_circleClaude Desktop & Claude Code
- check_circleCursor IDE & VS Code
- check_circleGemini CLI & OpenAI SDK
Async Jobs
Every operation runs serverless with real-time progress tracking.
Version Control
Immutable history. Restore to any version. Branch for experiments.
Team Collaboration
Shared datasets with role-based access, audit logs, and edit attribution across every version.
Clone & Sample
Create exact copies or stratified random subsets of any dataset.
Merge Datasets
Combine multiple datasets with automatic class reindexing and mapping.
Augment
Annotation-aware flips, rotations, brightness, contrast, blur, and noise.
Clean & Validate
Remove corrupted or unlabeled items, check dataset integrity per split.
Label Operations
Rename, remove, replace, or split classes across the entire dataset.
Resplit
Redistribute train/val/test with custom ratios while preserving class balance.
Deduplicate
Find and remove duplicates by checksum, filename, or file path.
Count & Report
Label distribution analysis with CSV/JSON export and per-split breakdown.
Import From
Anywhere
Search, preview, and import datasets from the largest public repositories — or upload your own files directly.
- hub
Hugging Face
Search 200K+ datasets with tag filtering and streaming for large files
- public
Kaggle
Browse 100K+ datasets with size filtering and credential-based search
- precision_manufacturing
Roboflow
Connect workspaces and import projects with annotations and class maps
- upload_file
Local Upload
Drag-and-drop ZIP, TAR, images, or annotations with auto-format detection
> Search Hugging Face for "COCO detection"
Found 847 datasets matching query
─────────────────────────────────
coco-2017 │ 118K images │ detection
coco-2014 │ 82K images │ detection
lvis │ 164K images │ detection + seg
> Import coco-2017 --splits train,val
● Downloading train (118,287 images)
● Processing annotations.json → 860K boxes
● Hashing content-addressable storage
✓ Imported as "coco-2017" · v1 created
> Augment coco-2017 --flip --rotate --factor 2
● Running augment_dataset
strategy: horizontal_flip, rotation_90
factor: 2× per image
✓ v2 created · 236,574 images · 1.72M boxesExport &
Download
Convert between any format. Filter by split, version, and class. Save to your Export Library for re-download.
Granular Filters
Select specific splits, filter by class, choose any version before exporting.
Export Library
All exports saved with status tracking. Re-download anytime before expiration.
Serverless Batch
Large exports run asynchronously with real-time progress and retry on failure.