walter0807 / mocanet.pytorch Goto Github PK

Hi, could you share the pretrained model for test？

Hi, I am trying to convert the trained model to TorchScript. However, it seems that this function does not defined anywhere.

Could you take a look at this ?

Thank you very much !

After training, I want to use 'calculate_metrics_3d.py' to directly evaluate the trained model or transform .npy file into video file. I meet the following wrong, it seems that the 'visualize' module is not included in the 'scripts' module.

command:
python scripts/exp/calculate_metrics_3d.py --in_dir ./test/ --render

result:
Traceback (most recent call last): File "scripts/exp/calculate_metrics_3d.py", line 7, in <module> from scripts.visualize.visualize_keypoints_npy import render_preview ModuleNotFoundError: No module named 'scripts.visualize'

Hi Walter0807,sorry for the bother. Thank you for releasing the code!
I trained mocanet model by python train.py --config configs/mocanet.yaml.
After 'test.py', I could get a series of .npy(2D & 3D position) which from ANDROMEDA to PUMPKINHULK.
The questions are as follows:

1. Could I directly used the pretrained model to infer motion、body、view from a single in-the-wild video?
2. If not, what is the requirement of in-the wild video? Should it come from a same demonstrator body structure?
3. If yes, I want to use the canonical structure to predict 3D position, could this canonical structure be general to in-the-wild video?

I can get the UV map using Densepose but I need the body coordinates. Can u share the preprocessing code?

Good morning,

Thank you for the paper, it looks really interesting and I would like to try it out. However, when I try training your model follow the tutorial in your README, this line of code

python train.py --config configs/mocanet.yaml

will throw an error:

Traceback (most recent call last):
  File "/home/xana/code/mocanet.pytorch/train.py", line 111, in <module>
    train_with_config(config, opts)
  File "/home/xana/code/mocanet.pytorch/train.py", line 43, in train_with_config
    train_loader = get_dataloader("train", config)
  File "/home/xana/code/mocanet.pytorch/lib/data.py", line 31, in get_dataloader
    dataloader = DataLoader(dataset, shuffle=True,
  File "/home/xana/mambaforge/envs/ubi/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 344, in __init__
    sampler = RandomSampler(dataset, generator=generator)  # type: ignore[arg-type]
  File "/home/xana/mambaforge/envs/ubi/lib/python3.10/site-packages/torch/utils/data/sampler.py", line 107, in __init__
    raise ValueError("num_samples should be a positive integer "
ValueError: num_samples should be a positive integer value, but got num_samples=0

I put the dataset inside data/mixamo.

Could you provide me some insight to fix this ?

Thank you very much !

Hello, I would like to ask what is the right input to feed into the model. When extract the skeleton from densepose, the result is in the range of x in [0, width] and y in [0, height]. But the npy in the data section has range from [-2, 2]. Could you provide some instructions (scaling ratio, translation, etc ) how to transform into the right range to feed in the model ?

Thank you very much !