Researchinstance segmentationdepth estimationunsupervised learninghigh speed imaging
Single-Camera Framework Estimates Bubble Depth Proxy
8.1
Chaitanya Nayak and colleagues (submitted Jan 29, 2026) introduce a machine-learning framework that detects bubbles and estimates their depth using a single 20 kHz high-speed camera with 3 µm resolution. The method combines unsupervised clustering to create pseudo-labels with a small set of in-plane annotations to train a semi-supervised model, producing continuous depth-proxy scores and robust instance segmentation with AP 0.818.
Key Points
- 1Develops semi-supervised pipeline using unsupervised clustering and pseudo-labels for bubble depth estimation.
- 2Achieves continuous depth-proxy scores from single 20 kHz camera, handling deformed and overlapping bubbles.
- 3Enables depth-aware, robust instance segmentation with AP 0.818 and Precision 0.901, low FPR.
Scoring Rationale
High practical novelty and usability drive score, limited by preprint status and relatively narrow experimental scope.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
