STBMP

We appreciate your interest in our work, but we do apologize for the "Motion Increments" issue. After reviewing our code, we found that the implementation in our code indeed implicitly included information about the first frame of future motion, resulting in the mistake. In mid-July, we formally reported this issue to ACM MM23 and requested the withdrawal of our paper. We sincerely apologize for any inconvenience and misleading this may have caused to you, other readers, and our community. We hope that despite this error, other aspects of the research presented in the paper may still prove valuable to your work.

Best wishes with your research!

Framework

Dependencies

DataSet

Human3.6m in exponential map can be downloaded from here. Directory structure:

H3.6m
|-- S1
|-- S5
|-- S6
|-- ...
`-- S11

CMU mocap was obtained from the repo of ConvSeq2Seq paper. Directory structure:

cmu
|   |-- ...
|-- train
`-- test

3DPW from their official website. Directory structure:

3dpw
|-- imageFiles
|   |-- ...
`-- sequenceFiles
    |-- test
    |-- train
    `-- validation

Training commands

All the running args are defined in opt.py. We use following commands to train on different datasets and representations.

• Train on Human3.6M:

To train a model of short-term prediction task, run

CUDA_VISIBLE_DEVICES={GPU_ID} python main_h36m_3d.py --t_input_size=66  --s_input_size=20 --input_n=10 --output_n=10 --dct_n=20  --is_load=False 

To train a model of long-term prediction task, run

CUDA_VISIBLE_DEVICES={GPU_ID} python main_h36m_3d.py --t_input_size=66  --s_input_size=50 --input_n=25 --output_n=25 --dct_n=50  --is_load=False

• Train on CMU-MoCap:

  CUDA_VISIBLE_DEVICES={GPU_ID} python main_cmu_3d.py --t_input_size=75  --s_input_size=35 --input_n=10 --output_n=25 --dct_n=35  --is_load=False

• Train on 3DPW:

  CUDA_VISIBLE_DEVICES={GPU_ID} python main_3dpw_3d.py --t_input_size=69  --s_input_size=40 --input_n=10 --output_n=30 --dct_n=40  --is_load=False

Test commands

Modif --is_load from "False" to "True" after the above training commands.

Acknowledgments

Some of our code was adapted from SPGSN.

Licence

This project is licensed under the terms of the MIT license.

This readme file is going to be further updated.