Silk Weaving

Description

In a jacquard loom workshop in Krefeld, Germany, the gestures of a skilled silk weaver were captured. The expert’s height was 168 cm, and his weight was 62 kg. The creation of the punch cards was recorded for one hour. Due to the complexity and length of the tasks, the wrapping and preparation of the silk beams were recorded only once, taking about four hours to record. Next, the expert was continuously recorded weaving with looms of three different sizes. The recording only stopped when the weaver switched to a different loom.

The task of waiving with a loom was segmented into three main gestures (G01G02, and G03). In the first, the expert pushes the pedal down with his right leg at the same time that he pushes away the threads with his left hand. In the second, by controlling the shuttle that passes the thread horizontally with the right hand, he sends the shuttle to the other side with a quick pulling gesture. Finally in a third gesture, he pulls back the threads with the left hand while simultaneously releasing the pedal with the right leg. This process is repeated up to the end of the piece.

Name convention and segmentation

Silk weaving (SW*):

  • The creation of the punch cards (SWPC).
  • Preparation of the beam (SWPB).
  • Wrapping of the beam (SWWB).
  • Jacquard weaving with small loom (SWSL).
  • Jacquard weaving with medium size loom (SWML).
  • Jacquard weaving with large loom (SWLL).
Download SLW

Password: mo_worX5slw*hc

Metrics

Gesture recognition

Method Motion descriptors Number of classes Accuracy F1-score
HMMs Joint angles from all 52 IMUs 3 0.954 0.943
HMMs Joint angles from: RFA and H 3 0.620 0.610
HMMs Joint angles from: RSH1, LSH1, HE, LUL, RL 3 0.909 0.892
Hidden Markov Models (HMMs).

Motion generation

Method MAE RMSE Average U1
KF-GOM 14.267 28.787 0.390
KF-RGOM 9.207 22.868 0.201
VAE-RGOM 0.115 0.467 0.043
ATT-RGOM 0.246 0.721 0.048
* Mean Absolute Error (MAE), Root Mean Squared Error (RMS), and Theil Inequality Coefficient (U1); KF-GOM, KF-RGOM, VAE-RGOM, and ATT-RGOM.