Three estimators × three ground-truth datasets
Every method evaluated on every dataset with force-sensor labels, predicted vs ground truth per dataset — so dataset quality is controlled within each row, and differences between panels are differences between methods.
↖ how the method is designed overview gallery 中文The matrix (Spearman ρ, predicted vs F/T ground truth)
| dataset (gel type) | Ours — physics, 0 training frames |
FEATS U-net (trained: marker gel) |
FeelAnyForce (trained: markerless) |
|---|---|---|---|
| FEATS val (marker) | 0.74 | 0.96 · in-domain | 0.43 |
| FoTa cnc_Mini (markerless) | 0.43 (0.65 non-edge) | 0.07 | 0.83 |
| GlowTact (markerless) | 0.63 | 0.04 | 0.90 |
The pattern is the finding: each network dominates its own gel domain and collapses outside it (FEATS 0.96 → 0.04–0.07; FeelAnyForce 0.90 → 0.43), while the physics pipeline is the only estimator that works everywhere (0.43–0.74) — it never sees training data, so it has no domain to leave.
FEATS dataset — marker-dot gel
In-domain, the FEATS U-net is excellent (ρ=0.96) — the negative results elsewhere are domain effects, not a weak model. FeelAnyForce, markerless-trained, degrades on the dotted gel (0.43): the same knife cuts both ways.
FoTa cnc_Mini — markerless gel
Hard conditions: only 4 contact-free frames, 62% of presses near the pad border. Our edge-filtered ρ is 0.65; FeelAnyForce reaches 0.92 non-edge.
GlowTact — markerless gel, cleaned
Friendliest ground truth (centred presses, 10 free frames, 0–20 N). Per-family our ρ reaches 0.93; the pooled gap to FeelAnyForce is mostly object-dependent volume→force gain, which a single physical scale cannot capture but 200K supervised frames can.
React — no ground truth, so: do independent methods agree?
On the dataset we actually care about, the two surviving estimators — physics (zero training) and FeelAnyForce (200K frames) — agree at ρ=0.91, and FeelAnyForce reads ≈0 N on every frame the physics pipeline calls contact-free. Neither can copy the other's mistakes.
Takeaway
For labelling React (markerless Mini): FeelAnyForce as the primary labeller, the physics pipeline as an independent audit, disagreement rows flagged. For any new gel or sensor where no trained model matches the domain: the physics pipeline is the only option that works out of the box — and its FEATS-dataset score (0.74) shows what it does on a domain nobody tuned it for.