InterGen: Diffusion-based Multi-human Motion Generation under Complex Interactions
Paper • 2304.05684 • Published
Inference-only artifact for the InterCLIP evaluator released with InterGen. It embeds a caption and two synchronized InterHuman-262 motion tracks, and is the standard evaluator used by InterGen and InterMask on InterHuman.
InterHuman262EvaluatorThe artifact contains only model.safetensors and protocol metadata. Optimizer,
trainer, callback, and Lightning state are excluded. OpenAI CLIP is used only
for tokenization; all learned InterCLIP text and motion weights are contained in
the SafeTensors file.
from motius.evaluation.evaluators import InterHuman262Evaluator
evaluator = InterHuman262Evaluator.from_pretrained(
"ZeyuLing/motius-evaluator-interhuman-interclip",
device="cuda",
)
metrics = evaluator.evaluate_npz("gt.npz", {"InterGen": "pred.npz"})