DropleX

Companion code for "DropleX: Liquid Sensing on Tablet Touchscreens" (IMWUT 2026).

DropleX is the first system that enables liquid sensing using the capacitive touchscreen of commodity tablets with no hardware modification required. Our system is able to detect microliter-scale liquid samples, and perform through-container measurements for liquid analysis.

---

Hardware

Samsung Galaxy Note 10.1 (2014 edition) with Atmel maXTouch MXT1066T2 (52 × 32 electrode grid, 4.2 mm pitch). Raw delta values are read via mxt-app (open-source CLI, requires root).

Root access is required. A standard (non-rooted) tablet will not work.

---

Structure

• firmware/ — Android app that streams live capacitance data and renders it as a real-time heatmap.
• data/ — Raw per-frame delta CSV files organized by session (session_<name>/). Each session has a baseline frame (deltas_0.csv) and subsequent measurement frames.
• regions/ — Extracted sensor-region tensors (.npz) used directly by training scripts.
• tablet_session_visualizer.py — Desktop tool to play back recorded sessions as an animated heatmap. Click a region to extract and save it to regions/.
• train_*.py — Training scripts for each experiment:
  • train_liquid_classifier_rf.py — liquid type (tap / DI / ethanol)
  • train_coke_spiking_classifier.py — Coke ethanol-spiking level
  • train_wine_classifier.py — wine adulteration
  • train_nacl_classifier.py — NaCl concentration
• example_model_inference.py — Load a bundled model and run a prediction.
• my_models/ — Pre-trained CNN (Coke/ethanol spiking), available as .pth.
• ref_normal.csv — Reference capacitance map snapshot at state of rest.

---

Quickstart

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

# Visualize a session and extract sensor regions
python tablet_session_visualizer.py --f data/session_coke_unadulterated

# Run the bundled pre-trained model
python example_model_inference.py --model my_models/coke_spiking_multiclass_*.pth --sample

# Reproduce a paper experiment
python train_coke_spiking_classifier.py \
  --classes unadulterated,ethanol10,ethanol30,ethanol50,ethanol80 --epochs 50

All training scripts accept --help for the full option list.

---

Included Data

Sessions in data/ cover four experiments: Coke ethanol-spiking, wine adulteration, liquid type in plastic cups, and NaCl concentration. Matching .npz region files are in regions/.

---

Citation

BibTeX will be added once camera-ready proceedings are published.