ai-tools/roboflow

6 notes, read out of this brain and free to use. Each one was extracted from a source and is re-checked against its exam.

Roboflow Inference server local deployment

To deploy Roboflow Inference locally, first install Docker on your machine. Then run: pip install inference inference-cli inference-sdk && inference server start. The inference server will be available at http://localhost:9001.

Running object detection inference with Roboflow

Use InferenceHTTPClient from inference_sdk to run inference on images. Create a client with api_url='http://localhost:9001' and api_key set to your Roboflow API key. Call client.use_model(MODEL_ID) to select a model, then client.infer(image) to run inference. The predictions object contains 'time', 'image' (with width and height), and 'predictions' list with objects containing x, y, width, height, confidence, class, and class_id.

Save Roboflow predictions to Supabase

Create a Supabase client with create_client(url, key) from the supabase package. Insert predictions with: supabase.table('predictions').insert({"filename": image, "predictions": predictions}).execute(). Query with: supabase.table('predictions').select("predictions").filter("filename", "eq", image).execute().

CLIP embeddings endpoint URL

Roboflow Inference provides HTTP endpoints for CLIP embeddings: /clip/embed_image for image embeddings and /clip/embed_text for text embeddings. Both endpoints accept an api_key query parameter. For local inference, use http://localhost:9001. For cloud inference, use https://infer.roboflow.com.

Roboflow model training and deployment

Models can be trained on Roboflow's platform using end-to-end tools for data management, annotation, and deployment. Custom model weights can also be uploaded for deployment. All models have an infinitely scalable API and can be run locally.

CLIP embedding API format (text and image)

The CLIP embedding API supports both text and image embeddings via POST requests with an API key. **Text Embeddings:** POST to `/clip/embed_text?api_key={API_KEY}` with JSON payload containing a 'text' key with the text string. The response is JSON with an 'embeddings' key containing a list of embedding vectors. **Image Embeddings:** POST to `/clip/embed_image?api_key={API_KEY}` with JSON payload containing an 'image' object with 'type' set to 'base64' and 'value' set to the base64-encoded image bytes. The response is JSON with an 'embeddings' key containing a list of embedding vectors. Both endpoints return responses in the same format with embedding vectors in the 'embeddings' key.

Give your agent this brain