Documentation
GANimator
Motion Generation

Motion Generation

Test & Inference

    • data.yaml
    • main.yaml
    • model.yaml
    • metrics.yaml
    • util.yaml

    • Metrics

    Here we evaluate the trained GANimator using metrics from GANimator and SinMDM. We measure the local and global diversity of the generated motions, as well as their quality in terms of plausibility (fid) and coverage:

    metrics.yaml
    # @package _global_

_moai_:
  _definitions_:
    _collections_:
      _metrics_:
        features:
          coverage:
            pred: [gen_feats]
            gt: [gt_feats]
            _out_: [coverage]
          gdiv:
            pred: [gen_feats]
            gt: [gt_feats]
            _out_: [ganimator_gdiv]
          ldiv:
            pred: [gen_feats]
            gt: [gt_feats]
            _out_: [ganimator_ldiv]
          mdm_gdiv:
            pred: [motion_embed]
            gt: [motion_embed_gt]
            _out_: [mdm_gdiv]
          mdm_ldiv:
            pred: [clips_embeds]
            gt: [clips_embeds_gt]
            _out_: [mdm_ldiv]
          fid:
            pred: [motion_embed]
            gt: [motion_embed_gt]
            _out_: [fid]

    To run the GANimator evaluation use the following command:

    python -m moai run test /path/to/ganimator/conf/run/main.yaml --config-dir ./conf npz_filename=%input_npz_file_name% model_ckpt=%path/to/trained/GANimator/checkpoint.ckpt% +mdm_ckpt=%path/to/SinMDM/t2m/text_mot_match/model/finest.tar%

    Export

    The trained GANimator is able to generate variations of the learned motion, i.e., the same motion base but small variations in the high-level features, by sampling multiple codes from a Gaussian distribution. The generated motions can be exported in .bvh format using BVH exporter:

    metrics.yaml
    exporters:
  bvh:
    parents: [joint_parents]
    position: [ik.root_position]
    rotations: [euler]
    offsets: [joint_offsets]
    names:
      - 'Hips' # 0
      - 'LeftUpLeg' # 1
      - 'LeftLeg' # 2
      - 'LeftFoot' # 3
      - 'LeftToeBase' # 4
      - 'LeftToe_End' # 5
      - 'RightUpLeg' # 6
      - 'RightLeg' # 7
      - 'RightFoot' # 8
      - 'RightToeBase' # 9
      - 'RightToe_End' # 10
      - 'Spine' # 11
      - 'Spine1' # 12
      - 'Spine2' # 13
      - 'Neck' # 14
      - 'Head' # 15
      - 'LeftShoulder' # 16
      - 'LeftArm' # 17
      - 'LeftForeArm' # 18
      - 'LeftHand' # 19
      - 'RightShoulder' # 20
      - 'RightArm' # 21
      - 'RightForeArm' # 22
      - 'RightHand' # 23

    To generate new motions use the following command:

    python -m moai run test conf/run/main.yaml model_ckpt=%path/to/trained/GANimator/checkpoint.ckpt% mdm_ckpt=%path/to/SinMDM/t2m/text_mot_match/model/finest.tar% out_name=%exported_file_name% export_dir=%path/to/generated/motions%
    Last updated on
    Model ArchitectureGANimator