CATVISH GUIDE
Inference Flow Editor
The Flow Editor (BETA) is a visual programming environment for building computer vision pipelines. It enables you to chain together models, logic, and external actions without writing code.
Visual Concepts
The editor is node-based. Data flows from left to right.
- Nodes: Blocks that perform a specific action (e.g., "Detect Objects", "Crop Image", "Send Alert").
- Edges: The connecting lines that carry data (Images, JSON) between nodes.
- Pins: Input and Output points on a node.
Common Nodes
Input Nodes
Sources data into the pipeline.
Camera Source: Reads from Webcam or RTSP URL.Image Folder: Iterates through files in a directory.
Model Nodes
Executes AI inference.
YOLO Detector: Loads a Catvish trained model (`.pt` or `.onnx`) and outputs bounding boxes.Classifier: Outputs class probabilities for a cropped image.
Logic Nodes
Filters and processes results.
Filter Detection: Filters boxes by Class Name or Confidence Threshold.Crop Objects: Crops the detected area from the original image for downstream processing.
Example Pipeline
GOAL:Detect "Person" and save the image if confidence > 90%.
Camera Source
→YOLO (v8m.pt)
→Filter (class="person", conf>0.9)
→Save to Disk