> For the complete documentation index, see [llms.txt](https://docs.augelab.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.augelab.com/function-blocks/blocks-reference/detections-shapes/detectors/custom-cnn-model.md). # Custom CNN Model This function block lets you load a TensorFlow-based image classification model and run predictions on incoming images. It is intended for users who want to apply a custom trained convolutional neural network to classify images in a visual workflow. ## đŸ“Ĩ Inputs `Image Any` The image that will be classified by the loaded model. ## 📤 Outputs `Class Index (Detected)` The numeric index of the predicted class. `Class Name (Detected)` The human-readable class name (if the model provides class labels). `Detection Result (Raw Output)` The raw model output or scores produced by the classifier. ## 🕹ī¸ Controls `Load Model` Opens a file dialog to select and load a model file. Supported model types include standard TensorFlow models and TFLite models. `Classes` A text area that displays the class list found inside the loaded model (if available). ## 🎨 Features * Simple model import: Load a trained TensorFlow or TFLite model using a single control. * Class label preview: The block shows class names extracted from the model so you can verify labels before running. * Immediate inference: When a model is loaded, incoming images are classified and the results are output for downstream use. * Outputs for automation: Both numeric index and human-readable class name are available, plus raw model outputs for advanced handling. ## ⚙ī¸ How it runs * Use the `Load Model` control to import a model file from disk. * Once a model is loaded and verified, the block accepts images via the `Image Any` input. * For each incoming image, the block performs inference and provides the predicted class index, optional class name, and the raw prediction values on the outputs. ## 📝 Usage Instructions 1. Click the `Load Model` button and choose your model file. Confirm that the class list appears in the `Classes` display to ensure the model is correctly loaded. 2. Connect any image source to the `Image Any` input (for example `Camera USB`, `Load Image`, or `Stream Reader`). 3. Use the block outputs to route classification results to visualization, logging, or decision logic blocks. ## 💡 Tips and Tricks * Preprocess images to match your model's expected input size with `Image Resizer` or `Image Resize` before feeding them to this block. This improves prediction consistency. * For visual verification, connect this block's image source to `Show Image` so you can preview what the model sees. * Overlay classification results onto images using `Write Text On Image` or `Draw Result On Image` to create easy-to-understand visual outputs for operators. * Save classification events with `Image Logger` or export numeric results to files using `CSV Export` for later analysis. * Use `Image ROI Select` or `Image ROI` to crop to relevant areas before classification if your model expects objects at a fixed position. * Combine with `Is None` or `Data Memory` to handle missing frames or freeze outputs when needed. ## 🛠ī¸ Troubleshooting * No classes shown after loading: Ensure the model contains class label metadata. If the class list is empty, the block will still output numeric indices but no human-readable names. * Model not loading: Verify you selected a supported model file and that the file path is accessible. If a TFLite or TensorFlow file is available, choose the appropriate file. * Unexpected predictions: Confirm input image size, color channels, and preprocessing match what the model was trained on. Use `Image Resize`, color conversion blocks, or preprocessing steps upstream to match training conditions. * Slow inference: Consider resizing images smaller with `Image Resizer` or using a lighter model variant before real-time deployment.