Experiments.md

Experiments

Our PointHMR can be run by following command

Pretraining scheme

We conduct large-scale training on multiple 2D and 3D datasets, including Human3.6M, COCO, MUCO, UP3D, MPII. During training, it will evaluate the performance per epoch, and save the best checkpoints.

sh script/pretrain.sh

• nproc_per_node: Number of GPU.

• train_yaml: Path of training datasets (Human3.6M, COCO, MUCO, UP3D, MPII).

• val_yaml: Path of valiadation dataset (Human3.6M).

Finetuning scheme

We follow prior works that also use 3DPW training data. During training, it will evaluate the performance per epoch, and save the best checkpoints. We fine-tune on 3DPW training set with the checkpoint from the pretraining scheme by the following command.

sh script/finetuning.sh

• nproc_per_node: Number of GPU.

• train_yaml: Path of training dataset (3DPW).

• val_yaml: Path of validation dataset (3DPW).

• resume_checkpoint: Path to the specific checkpoint (checkpoint from pretraining) for resume training.

Evaluation Human3.6M

We evaluate our PointHMR on the Human3.6M dataset by the following command.

sh script/evaluation_h36m.sh

• nproc_per_node: Number of GPU.

• val_yaml: Path of validation dataset (Human3.6M).

• resume_checkpoint: Path to the specific checkpoint for evaluation.

Evaluation 3DPW

We evaluate our PointHMR on the 3DPW dataset by the following command.

sh script/evalutation_3dpw.sh

• nproc_per_node: Number of GPU.
• val_yaml: Path of validation dataset (3DPW).
• resume_checkpoint: Path to the specific checkpoint for evaluation.