SentiSight SDK: Complete Guide to Getting Started
1. What SentiSight SDK is
- Purpose: Computer-vision toolkit (image classification, object detection, instance segmentation, pose estimation, text recognition, background removal, similarity search).
- Provider: SentiSight / Neurotechnology.
2. Quick prerequisites
- API account on SentiSight.ai (register to get API token).
- Images and labeled data for custom training (if training models).
- Development environment: Python/JavaScript/Java/C# toolchain as needed.
- HTTP client (curl, requests, axios) or SDK libraries where available.
3. Core workflows
- Create a project (via web UI or REST API).
- Upload images (file, URL, or Base64).
- Label images (manual or assisted — labeling by similarity).
- Train model (SentiSight handles training; monitor training statistics).
- Make predictions (use pre-trained models or your trained model).
- Download model for offline use (optional).
4. Using pre-trained models (fast path)
- Endpoint pattern: https://platform.sentisight.ai/api/pm-predict/{model_name}/
- Common model names: General-classification, Places-classification, NSFW-classification, Object-detection, Instance-segmentation, Text-recognition, Pose-estimation, Background-removal.
- Authentication: set header
X-Auth-token: YOURTOKEN. - Send image as binary with Content-Type: application/octet-stream (or use image URL/Base64 per API docs).
- Example curl:
Code
TOKEN=“your_token” MODEL=“General-classification” curl -H “X-Auth-token: \(TOKEN" --data-binary @"image.jpg" \-H "Content-Type: application/octet-stream" \ -X POST "https://platform.sentisight.ai/api/pm-predict/\)MODEL”
5. Training a custom model
- Create a labeled project (classification, detection, or segmentation).
- Upload and assign sufficient examples per label (more samples → better accuracy).
- Start training via UI or API; monitor metrics on “View training statistics”.
- Export/download trained model for offline deployment if needed.
6. Making predictions with your trained model
- Use prediction endpoint: https://platform.sentisight.ai/api/predict/{projectId}/{modelName}
- Authenticate with
X-Auth-token. - Supported inputs: file upload, image URL, Base64.
- Results available in JSON; downloads as CSV/JSON/ZIP depending on task.
7. Offline usage
- From training statistics page, click “Download model” to get an offline package.
- Host your own REST API server to serve predictions locally.
8. Useful resources
- Official user guide: https://www.sentisight.ai/user-guide/
- Pre-trained models docs: https://www.sentisight.ai/user-guide/using-pre-trained-models/
- API/OpenAPI (Swagger): https://app.swaggerhub.com/apis-docs/SentiSight.ai/sentisight
9. Practical tips
- Use labeling-by-similarity to speed annotation.
- Start with pre-trained models for prototyping.
- Keep a validation set to evaluate real accuracy.
- For production, prefer offline model hosting if latency or data privacy is critical.
If you want, I can generate sample code (Python, JavaScript, or curl) for your chosen task (predict, train, or download model).
Leave a Reply