The dataset does not contain attribute faces.

[YuRongHao2233]

When I want to visualize the human-object-motion, it hints me there is no attribute faces in the file person1_poses.npz

[OctopusWen]

@YuRongHao2233 I'm sorry. However, since we used SMPLX to store the human pose, the "faces" attribute is not provided. You can use SMPLX inference to get the faces. More details can be found in the docs/file_definitions.md.

[YuRongHao2233]

So do I need to replace the faces in your visualization code with the faces provided in this code? @OctopusWen

[YuRongHao2233]

And I found that some data objects are not synchronized with people's movements, and some objects such as (board007_m.obj) lack faces

[YuRongHao2233]

ModuleNotFoundError: No module named 'data_processing'，and when I try to run forcasting task, I found there is a lack of module, I can not find it in your code.

[leolyliu]

ModuleNotFoundError: No module named 'data_processing'，and when I try to run forcasting task, I found there is a lack of module, I can not find it in your code.

Hi YuRongHao2233:

Thank you for pointing out this issue. We have uploaded the data_processing module under ./benchmarks/data_processing. Please refer to ./benchmarks/data_processing/README.md to install the environment.

We will clean up the benchmark code and continuously fix the potential bugs in its environment setup. The algorithms in the code are aligned with those described in our arxiv paper.

Best

[leolyliu]

And I found that some data objects are not synchronized with people's movements, and some objects such as (board007_m.obj) lack faces

(1) For the synchronization problem, could you please provide the sequence names of the data?

(2) For the problem of object models, several board models are broken before, we have fixed them and re-uploaded the object_model.zip on OneDrive. Besides, we will upload high-precision object models soon.

[leolyliu]

So do I need to replace the faces in your visualization code with the faces provided in this code? @OctopusWen

Yes, exactly. We have updated visualize_human_object_motion.py and its usage. Thank you for the feedback.

[YuRongHao2233]

ModuleNotFoundError: dataset_statistics，and when I try to run forcasting task, I found there is a lack of module, I can not find it in your code, too. And for some .py file, there are some bugs in them, I have fixed them, now I wish you upload the dataset_statistics

[YuRongHao2233]

And I found that some data objects are not synchronized with people's movements, and some objects such as (board007_m.obj) lack faces

(1) For the synchronization problem, could you please provide the sequence names of the data?

(2) For the problem of object models, several board models are broken before, we have fixed them and re-uploaded the object_model.zip on OneDrive. Besides, we will upload high-precision object models soon.

Of course, I will give you the name of anomaly files I find.

[leolyliu]

ModuleNotFoundError: dataset_statistics，and when I try to run forcasting task, I found there is a lack of module, I can not find it in your code, too. And for some .py file, there are some bugs in them, I have fixed them, now I wish you upload the dataset_statistics

We have uploaded dataset_statistics, including a train-test split. Since the benchmark algorithms are evaluated on our preprocessed data ("/data.npz"), and the preprocessed data is too large to upload, we will modify the data processing modules to fit the current dataset.

[YuRongHao2233]

So can I run a benchmark on a synthetic dataset

[leolyliu]

So can I run a benchmark on a synthetic dataset

Yes. Before running the motion forecasting and synthesis benchmarks, please modify benchmarks/motion_forecasting/MDM_Interdiff/interdiff/data/prepare_hho.py to align with the current data formats and run it. This code can generate <sequence_name>/data.npz. Or please wait up to two days to let us fix it. Sorry for the data format misalignment. The CORE4D-Synthetic shares similar formats with CORE4D-Real.

[leolyliu]

The benchmark code has been updated. We have fixed the data formatting issues.

So can I run a benchmark on a synthetic dataset

Yes. Before running the motion forecasting and synthesis benchmarks, please modify benchmarks/motion_forecasting/MDM_Interdiff/interdiff/data/prepare_hho.py to align with the current data formats and run it. This code can generate <sequence_name>/data.npz. Or please wait up to two days to let us fix it. Sorry for the data format misalignment. The CORE4D-Synthetic shares similar formats with CORE4D-Real.

[YuRongHao2233]

And I found that some data objects are not synchronized with people's movements, and some objects such as (board007_m.obj) lack faces

(1) For the synchronization problem, could you please provide the sequence names of the data?

(2) For the problem of object models, several board models are broken before, we have fixed them and re-uploaded the object_model.zip on OneDrive. Besides, we will upload high-precision object models soon.

20231002, 055 and 028 , they are obvious. And maybe other sequences are not synchronized

[YuRongHao2233]

How do I reconcile the data of ego with the named file of human-motion?

[OctopusWen]

How do I reconcile the data of ego with the named file of human-motion?

Hi @YuRongHao2233 , The timestamp of ego is provided in *.mp3 files coded in LTC format. And the timestamp of human-motion is supplied by timestamp.txt at the corresponding allocentric_RGBD_videos.

[YuRongHao2233]

How do I reconcile the data of ego with the named file of human-motion?

Hi @YuRongHao2233 , The timestamp of ego is provided in *.mp3 files coded in LTC format. And the timestamp of human-motion is supplied by timestamp.txt at the corresponding allocentric_RGBD_videos.

Hi， I mean how to match human-motion files to egocentric videos, I can't tell how they correspond at the moment, and I hope you can help me