Loading dataset...
Smith300DA3Dataset: loading 9 sessions
  dataset_20260512_161202: 123 frames loaded
  dataset_20260512_161418: 107 frames loaded
  dataset_20260512_161822: 123 frames loaded
  dataset_20260512_162057: 102 frames loaded
  dataset_20260512_162940: 138 frames loaded
  dataset_20260512_163220: 91 frames loaded
  dataset_20260512_163457: 97 frames loaded
  dataset_20260516_163657: 133 frames loaded
  extra_singleview_capture: 616 frames loaded
Smith300DA3Dataset ready: 62 eps, 1530 frames, 1468 samples, rgb=4.66 GB, depth=1.55 GB
  Train: 1395  Val: 73
Building model...
[INFO ] using MLP layer as FFN
Loading weights from local directory
Trainable: 25,932,928
Epoch 12:  79%|███████▉  | 138/174 [00:21<00:09,  3.60it/s, d=0.017, h=8.105, w_d=60, w_h=0.12]  
Epoch 0: val_h=11.0707  val_d=0.0478  val_pix=122.2px
  ✓ best (val_h=11.0707)
Epoch 1: val_h=10.8349  val_d=0.0382  val_pix=116.6px
  ✓ best (val_h=10.8349)
Epoch 2: val_h=10.5692  val_d=0.0342  val_pix=94.4px
  ✓ best (val_h=10.5692)
Epoch 3: val_h=10.2976  val_d=0.0341  val_pix=90.1px
  ✓ best (val_h=10.2976)
Epoch 4: val_h=10.0665  val_d=0.0325  val_pix=77.3px
  ✓ best (val_h=10.0665)
Epoch 5: val_h=9.7666  val_d=0.0293  val_pix=79.3px
  ✓ best (val_h=9.7666)
Epoch 6: val_h=9.5153  val_d=0.0300  val_pix=67.1px
  ✓ best (val_h=9.5153)
Epoch 7: val_h=9.2954  val_d=0.0286  val_pix=72.0px
  ✓ best (val_h=9.2954)
Epoch 8: val_h=9.1318  val_d=0.0293  val_pix=62.7px
  ✓ best (val_h=9.1318)
Epoch 9: val_h=8.8014  val_d=0.0281  val_pix=60.8px
  ✓ best (val_h=8.8014)
Epoch 10: val_h=8.7092  val_d=0.0295  val_pix=56.7px
  ✓ best (val_h=8.7092)
Epoch 11: val_h=8.3981  val_d=0.0305  val_pix=64.3px
  ✓ best (val_h=8.3981)