processing_viz

Interactive Processing sketch that visualizes a small CNN: four conv layers followed by two MLP layers, fed by a 32×32 drawing canvas. Weights/biases are pre-exported from training and loaded at startup; the sketch animates feature maps, filter movements, reshaping, and final class logits with a PeasyCam-controlled 3D view.

Requirements

• Processing 4.x
• PeasyCam library (install via Processing Contributions Manager)

Run

1. Open processing_viz.pde in Processing.
2. Ensure data/ is alongside the sketch (already included).
3. Run the sketch.
4. Draw a digit in the bottom-right canvas; the network forwards through convs → flatten → MLP → softmax and highlights the predicted class box.

Files

• processing_viz.pde – main sketch orchestrating inference + visualization.
• Conv2D.pde, MLP.pde, TensorVisualizer.pde, ConvVisualizer.pde, ReshapeVisualizer.pde, Box.pde, utils.pde – layer math and rendering helpers.
• data/*.txt – serialized weights/biases and test tensors.
• preview.png – static preview of the UI layout.

Notes

• Camera roll is suppressed for easier navigation; mouse over the draw canvas temporarily disables camera controls to avoid conflict with drawing.
• You can regenerate weights and replace data/*.txt if you retrain; keep shapes consistent.