3DPrintSaviour for Marlin + Dual Cameras

Objective

This fork adapts the original OctoPrint/Octolapse workflow to a custom Marlin-based printer that is connected directly over USB serial and monitored with two cameras.

The new workflow is:

1. run opens two cameras and captures a background frame for each one.
2. It listens to the Marlin serial console for layer markers such as M118 LAYER:12.
3. On every trigger, it captures frames from both cameras, computes the original NRMSE-based metrics per camera, and optionally runs spaghetti detection.
4. If either camera reports a configured failure condition, printcontrol.py sends Marlin pause G-code over serial.

If layer messages are not available yet, run can also fall back to timed captures.

• SERIAL_PORT to the printer port, for example COM5
• PAUSE_COMMANDS to the pause/stop G-code that matches the printer workflow
• CAMERAS so both camera sources, crop areas, rotation, and mirroring are correct
• TRIGGER_MODE to layer, timer, or hybrid

Recommended Marlin trigger setup

For best results, make your slicer emit a serial message at every layer change (didn't work well without this). A simple pattern is:

M118 LAYER:[layer_num]

That lets run capture images at repeatable layer boundaries instead of arbitrary time intervals.

Installation

Create a Python environment and install the runtime dependencies:

pip install numpy scipy scikit-image opencv-python pyserial matplotlib

For spaghetti detection, also download the Darknet weights file into ml_api/model/model.weights using the URL in ml_api/model/model.weights.url.

Running the monitor

The monitor needs to be started before beginning the print so capture index 000000.jpg becomes the empty-bed background reference:

python run --job-name test_part

Useful options:

python run --trigger-mode layer
python run --trigger-mode timer --capture-interval 10
python run --background-only --job-name calibration

Captured sessions are written under captures/<job-name>/ with this layout:

captures/
  test_part/
    cam0_top/
      000000.jpg
      000001.jpg
      output.log
    cam1_side/
      000000.jpg
      000001.jpg
      output.log
    combined/
      000000.jpg
      000001.jpg
    events.log

Detection logic

The score rules run per camera:

• Detachment: score > 1.0 and deviance > 1.0
• Breakage: score_diff > 0.2 and deviance_diff > 0.2
• Filament runout/clog: score < 0.2 and deviance < 0.2
• Spaghetti: Darknet model reports one or more spaghetti detections

The confirmation mode is controlled by FAILURE_CONFIRMATION_MODE in api_keys.py:

• any: pause when one camera confirms a failure
• all: pause only when both cameras confirm a failure

Utility scripts

• get_score.py calculates metrics for one saved frame
• data_visualisation.py plots one camera output.log
• gen_video.py builds a preview video from a camera or combined capture folder

Notes for custom printers

• Some Arduino-style boards reset when a serial connection opens. DISABLE_DTR_ON_CONNECT is included to reduce that risk, but behaviour still depends on your USB interface and firmware board.
• M25 is mainly for SD-card prints. If your printer is driven differently, adjust PAUSE_COMMANDS to match your Marlin setup.
• For the most reliable image comparison, park the toolhead out of frame at each trigger point or choose camera angles where the nozzle does not hide the part.