IMU4D model weights

Eight IMU4D generator weights for wearable-IMU motion, text, and scene inference. These are internal training versions. The public training scripts in the code repository provide a training recipe; they do not describe the exact training history of these weight files.

Model name	File	Task
pretrain	checkpoints/showo_pretrain_full/pytorch_model.bin	General motion and text
noise	checkpoints/showo_pretrain_full_noise/pytorch_model.bin	Noise augmented IMU
imuposer	checkpoints/showo_finetune_imuposer/pytorch_model.bin	IMUPoser
dipimu	checkpoints/showo_finetune_dipimu/pytorch_model.bin	DIP-IMU
ncsa	checkpoints/showo_finetune_ncsa_fix/pytorch_model.bin	NCSA capture
hiphi	checkpoints/showo_finetune_hiphi/pytorch_model.bin	HiPHI scenes
omomo	checkpoints/showo_finetune_omomo/pytorch_model.bin	OMOMO scenes
humoto	checkpoints/showo_finetune_humoto/pytorch_model.bin	HUMOTO scenes

Install the code and follow its README for dependencies, the included motion tokenizer, input format, and inference. For example:

python scripts/download_checkpoints.py pretrain
python scripts/infer.py --model pretrain \
  --input dataset_process/sample_data/LINGO_17992.pkl \
  --output exp/inference/lingo_example

The weights contain model parameters for inference. They do not include optimizer states or intermediate training snapshots. Processed evaluation data is in IMU4DData.