force recovery: methods, evaluation, debug log

results.html

@@ -66,10 +66,25 @@ better of the two. Nothing but the image→gradient step differs.</p>
 <div class="tablewrap"><table><thead><tr><th>dataset</th><th>n</th><th>LUT</th><th>calibration-free</th><th>shuffle floor</th><th>groups fitted</th></tr></thead><tbody>
 <tr><td>GelSight Mini, CNC presses, 0-20 N</td><td>605</td><td>0.993</td><td><b>0.996</b></td><td class='dim'>+0.067 / +0.023</td><td class='dim'>6/6</td></tr><tr><td>FoTa cnc_Mini</td><td>528</td><td>0.910</td><td><b>0.961</b></td><td class='dim'>+0.000 / +0.065</td><td class='dim'>6/6</td></tr><tr><td>FEATS (MARKER gel)</td><td>219</td><td><b>0.583</b></td><td>0.526</td><td class='dim'>-0.076 / +0.045</td><td class='dim'>3/6</td></tr><tr><td>Sparsh / Meta, 10 gel pads</td><td>407</td><td>0.929</td><td><b>0.948</b></td><td class='dim'>+0.166 / +0.116</td><td class='dim'>8/8</td></tr><tr><td>FeelAnyForce</td><td>1150</td><td>0.943</td><td><b>0.958</b></td><td class='dim'>+0.802 / +0.791</td><td class='dim'>19/32</td></tr></tbody></table></div>
 
+<figure><img src="assets/truncation.png" alt="truncated presses">
+<figcaption>What the row above excludes. A press is <b>truncated</b> when its
+contact core reaches an image border: part of the indentation is outside the
+frame, the free-boundary solve has a slope running off the edge and nothing on
+the far side to stop it, and the height ramps. 14.5% of truncated frames
+reconstruct deeper than the 4.25&nbsp;mm gel is thick, against 0%
+of frames imaged whole. Their depth is not identifiable from the image, so no
+reconstruction can fix them.</figcaption></figure>
+
+<p>Excluding them is what the headline row buys: on GelSight Mini, CNC presses, 0-20 N
+calibration-free scores ρ&nbsp;0.996
+on whole presses against
+0.863 once truncated frames are
+mixed in. They are also the minority — GelSight Mini, CNC presses, 0-20 N 605 of 2,786; FoTa cnc_Mini 528 of 2,499; FEATS (MARKER gel) 219 of 390; Sparsh / Meta, 10 gel pads 407 of 960; FeelAnyForce 1,150 of 5,410. <b>A uniform 2,000-frame
+evaluation does not exist</b>: no dataset here contains that many fully imaged
+presses.</p>
+
 <h3>All frames</h3>
-<p class="dim">Most capture grids are larger than the field of view, so this
-population is dominated by contacts whose extent is outside the image and whose
-depth is not identifiable from it.</p>
+<p class="dim">The same protocol without that exclusion, for reference.</p>
 <div class="tablewrap"><table><thead><tr><th>dataset</th><th>n</th><th>LUT</th><th>calibration-free</th><th>shuffle floor</th><th>groups fitted</th></tr></thead><tbody>
 <tr><td>GelSight Mini, CNC presses, 0-20 N</td><td>2786</td><td>0.660</td><td><b>0.863</b></td><td class='dim'>+0.026 / +0.019</td><td class='dim'>6/6</td></tr><tr><td>FoTa cnc_Mini</td><td>2499</td><td>0.401</td><td><b>0.491</b></td><td class='dim'>+0.060 / +0.047</td><td class='dim'>6/6</td></tr><tr><td>FEATS (MARKER gel)</td><td>390</td><td><b>0.775</b></td><td>0.685</td><td class='dim'>-0.003 / +0.067</td><td class='dim'>4/6</td></tr><tr><td>Sparsh / Meta, 10 gel pads</td><td>960</td><td>0.456</td><td><b>0.588</b></td><td class='dim'>+0.092 / +0.099</td><td class='dim'>8/8</td></tr><tr><td>FeelAnyForce</td><td>5410</td><td>0.850</td><td><b>0.862</b></td><td class='dim'>+0.640 / +0.636</td><td class='dim'>42/42</td></tr></tbody></table></div>
 
@@ -99,5 +114,5 @@ frames, mean difference 0.84&nbsp;N. <b>The published dataset still
 carries the LUT column</b>: switching it needs 36 episodes reprocessed.</p>
 
 <h2>Error analysis</h2>
-<figure><img src="assets/errors_cnc_mini_26.png" alt="cnc_mini_26 errors"><figcaption>GelSight Mini CNC — the ten worst held-out frames and, as a control, the five best. Relative error is |pred−true| over the dataset's force span (19.50&nbsp;N); median 1.0%, p90 4.5%, worst 23.6%.</figcaption></figure><figure><img src="assets/errors_cnc.png" alt="cnc errors"><figcaption>FoTa cnc_Mini — the ten worst held-out frames and, as a control, the five best. Relative error is |pred−true| over the dataset's force span (4.01&nbsp;N); median 3.0%, p90 10.4%, worst 29.6%.</figcaption></figure><figure><img src="assets/errors_feats.png" alt="feats errors"><figcaption>FEATS (marker) — the ten worst held-out frames and, as a control, the five best. Relative error is |pred−true| over the dataset's force span (34.11&nbsp;N); median 5.7%, p90 22.1%, worst 67.9%.</figcaption></figure><figure><img src="assets/errors_sparsh.png" alt="sparsh errors"><figcaption>Sparsh / Meta — the ten worst held-out frames and, as a control, the five best. Relative error is |pred−true| over the dataset's force span (0.98&nbsp;N); median 2.6%, p90 11.7%, worst 30.0%.</figcaption></figure><figure><img src="assets/errors_faf.png" alt="faf errors"><figcaption>FeelAnyForce — the ten worst held-out frames and, as a control, the five best. Relative error is |pred−true| over the dataset's force span (17.42&nbsp;N); median 2.5%, p90 10.7%, worst 44.6%.</figcaption></figure>
+<figure><img src="assets/errors_cnc_mini_26.png" alt="cnc_mini_26 errors"><figcaption>GelSight Mini CNC — the ten worst held-out frames and, as a control, the five best. Relative error is |pred−true| over the dataset's force span (19.50&nbsp;N); median 1.1%, p90 3.7%, worst 11.6%.</figcaption></figure><figure><img src="assets/errors_cnc.png" alt="cnc errors"><figcaption>FoTa cnc_Mini — the ten worst held-out frames and, as a control, the five best. Relative error is |pred−true| over the dataset's force span (4.01&nbsp;N); median 3.8%, p90 10.9%, worst 48.3%.</figcaption></figure><figure><img src="assets/errors_feats.png" alt="feats errors"><figcaption>FEATS (marker) — the ten worst held-out frames and, as a control, the five best. Relative error is |pred−true| over the dataset's force span (34.11&nbsp;N); median 5.7%, p90 22.1%, worst 67.9%.</figcaption></figure><figure><img src="assets/errors_sparsh.png" alt="sparsh errors"><figcaption>Sparsh / Meta — the ten worst held-out frames and, as a control, the five best. Relative error is |pred−true| over the dataset's force span (0.98&nbsp;N); median 3.1%, p90 12.0%, worst 30.0%.</figcaption></figure><figure><img src="assets/errors_faf.png" alt="faf errors"><figcaption>FeelAnyForce — the ten worst held-out frames and, as a control, the five best. Relative error is |pred−true| over the dataset's force span (17.43&nbsp;N); median 1.6%, p90 12.7%, worst 48.1%.</figcaption></figure>
 </div></body></html>