Hi, Thanks for your codes. I wonder why the results of "original version" are different from the paper and the pytorch version is a little better than the paper?

Thanks a lot!

Hi @weigq,

Thanks for the great implementation. I'm trying to train on SHG detections but the .zip file doesn't contain the test_2d_ft.pth.tar file.

Could you provide detection files or processing script for that?

I follow with https://github.com/weigq/3d_pose_baseline_pytorch/blob/a03094bb479770abc43dc5b87ad394b85bb6dcab/main.py#L226 to unnorm the output of network, and get tensor which has (51, ) shape. I guess it should be the coordinates like x0,y0,z0,x1,y1,z1.......x16,y16,z16 but when I draw all 17 point in 3d plot, it don't like a human skeleton shape,

I want to know how to visualize the output from the (51, ) array, and the index of these 51 nums.

Hi there,

Your code looks great. I was just wondering what is the main difference between your pre-processed dataset and h36m.zip from the original tensorflow repo.

Thanks

Hi, since the dataset provided here is normalized version, could anyone give me idea how to unnormalize 2D Skeleton? I already see this function and successfully unnormalize the 3D Skeleton, but how we use this code to unnormalize 2D Skeleton? I will really glad if someone could help. Also is there a function to do normalize? Thank you.

def unNormalizeData(normalized_data, data_mean, data_std, dimensions_to_use):

Hello,

The SH detections do not seem to be available anymore. Also in the main repo (https://github.com/una-dinosauria/3d-pose-baseline) there is no code available for hourglass detections. Does anyone happen to have the detection files, i.e. 'train_2d_ft.pth.tar' and 'test_2d_ft.pth.tar'? Or the code for the stacked hourglass model that was used to produce the results?

At the moment I am only able to run my experiments with the GT. It would be great if someone could help me with the SH detections :)

Cheers, Duncan

Thanks for the great repo! Your code is very elegant. Just a quick question. I am trying to plot the given 2d and 3d processed skeletons. However, I could not find the connection between joints in your repo. I am wondering if you could either provide the limb connections or the code for visualizing?

Thanks a lot!

dear @weigq, What is that?

optional arguments, you can access more details in opt.py

main.py [-h] [--data_dir DATA_DIR] [--exp EXP] [--ckpt CKPT] [--load LOAD] [--test] [--resume] [--action {all,All}] [--max_norm] [--linear_size LINEAR_SIZE] [--num_stage NUM_STAGE] [--use_hg] [--lr LR] [--lr_decay LR_DECAY] [--lr_gamma LR_GAMMA] [--epochs EPOCHS] [--dropout DROPOUT] [--train_batch TRAIN_BATCH] [--test_batch TEST_BATCH] [--job JOB] [--no_max] [--max] [--procrustes]

Hi, Wanted to point out a bug in your error calculation. The errors for all actions are simply averaged, without regard to the number of data points in each action. It should be a weighted mean. This yields slightly higher numbers.

Hi there,

Thanks again for sharing your code, its very clear.

I have a quick question about the error computation. You select the valid dimensions here which includes dimensions for the hip (which is not predicted) for. As I understand it, in the squared error computation three of the dimensions will always have a very low error e.g. they are predicted zero and the gt is zeros.

Is this the same way as the original TensorFlow code?

Thanks

@weigq Hello! I've run train_2d_ft.pth.tar and test_2d_ft.pth.tar on other models, whose performance is very poor. So I want to know if the train_2d_ft.pth.tar contains the same kind of joints and the same order of joints as it did on train_2d.pth.tar and train_3d.pth.tar ? Thanks!

Hi! I'm confused when visualizing the poses. For example, I get a numpy.ndarray from test_2d.pth.tar, which has the shape (32,) and represent a 2d pose. I guess it arrays as x1,y1,x2,y2, ..., so I add 0,0 at the front of pose, and try to connect them following the sequence of the original version , but it dosen't look like a human. The same result on another arrayx1,x2,...,y1,y2, ..., and 3d pose as well. I want to know whether I misunderstanding some thing in the process above. And what's the shape of normalized bbox from original picture? How to change the coordinates from pixel to the form we see now in the files? Thanks!!!

Could you please share the data/stat_3d.pth.tar ?

[jalal@goku 3d_pose_baseline_pytorch]$ python main.py --exp example

==================Options=================
{   'action': 'All',
    'ckpt': 'checkpoint/example',
    'data_dir': 'data/',
    'dropout': 0.5,
    'epochs': 200,
    'exp': 'example',
    'is_train': True,
    'job': 8,
    'linear_size': 1024,
    'load': '',
    'lr': 0.001,
    'lr_decay': 100000,
    'lr_gamma': 0.96,
    'max_norm': True,
    'num_stage': 2,
    'procrustes': False,
    'resume': False,
    'test': False,
    'test_batch': 64,
    'train_batch': 64,
    'use_hg': False}
==========================================

>>> creating model
/scratch3/3d_pose/humanpose/3d_pose_baseline_pytorch/src/model.py:11: UserWarning: nn.init.kaiming_normal is now deprecated in favor of nn.init.kaiming_normal_.
  nn.init.kaiming_normal(m.weight)
>>> total params: 4.29M
/scratch/sjn-p3/anaconda/anaconda3/lib/python3.6/site-packages/torch/nn/functional.py:52: UserWarning: size_average and reduce args will be deprecated, please use reduction='elementwise_mean' instead.
  warnings.warn(warning.format(ret))
>>> actions to use (total: 15):
[   'Directions',
    'Discussion',
    'Eating',
    'Greeting',
    'Phoning',
    'Photo',
    'Posing',
    'Purchases',
    'Sitting',
    'SittingDown',
    'Smoking',
    'Waiting',
    'WalkDog',
    'Walking',
    'WalkTogether']
>>>
>>> loading data
Traceback (most recent call last):
  File "main.py", line 276, in <module>
    main(option)
  File "main.py", line 75, in main
    stat_3d = torch.load(os.path.join(opt.data_dir, 'stat_3d.pth.tar'))
  File "/scratch/sjn-p3/anaconda/anaconda3/lib/python3.6/site-packages/torch/serialization.py", line 356, in load
    f = open(f, 'rb')
FileNotFoundError: [Errno 2] No such file or directory: 'data/stat_3d.pth.tar'

how to get the data train_2d_ft.pth.tar, test_2d_ft.pth.tar, train_2d.pth.tar,test_2d.pth.tar and train_3d.pth.tar ? Thank you so much If you can help me !