guohengkai/region-ensemble-network

Repository for Region Ensemble Network based Hand Pose Estimation

region-ensemble-network's Issues

Install pyrealsense

@guohengkai

Thanks for sharing your code. There is some errors when installed the pyrealsense, can you give me the operations in details?

regards,
weiguo

Windows specifics: set environment variable PYRS_INCLUDES to the rs.h directory location and environment variable PYRS_LIBS to the librealsense binary location. You might also need to have stdint.h available in your path.

手关节点做约束了吗？谢谢

你好 我想问一下 你做的研究手势关节点做了约束没？

Speed and GPU memory

Nice work!
i wonder how fast is the algorithm?
And how many GPU memories does this algorithm occupy?

Training codes using REN

Hi,
Thank you very much for sharing codes! I'm a new beginner for the framework of caffe, can you give more details on the training processing? I
regards,
weiguo

Can GAP Layer replace FC layer?

Hi thans yours work. I'm interest to your work. There is a question for me. Fc Layer param is so much. Can GAP Layer replace FC layer in your network?

Can you release the training code?

Thank you for your awesome research. I want to use your method to train my own model but I find that there is only the prediction code in the repo. Can you release the training code together? I will be very grateful for that.

Data augmentation

Hi, Thanks for your research and paper. I am trying to implement it in pytorch.

I would like to ask about the data augmentation (translation, scaling, and rotation).
If i am understanding it correctly, first you get the cube (150x150), then resize to 96x96, then normalize to -1,1. This is all done using _crop_image.

How did you apply the transformations after _crop_image?

Thanks for your help.

左右手都可以回归的很好吗？或者说只支持其中一只手

test code of live results from Kinect 2

hi, thx for you work. Could you please send the code about live results from Kinect 2 sensor to me. I'm dong some researchs on it and looking forward to it. My email is [email protected]

data preprocessing

HI~
Thank you for your great work!
Could you please explain how do you get the hand center for me?
I see you have given the hand center coordinate in /labels/*_center.txt

Thank you very much!

About DATA Preprocessing。

@guohengkai @xinghaochen THx!Your Paper is helpful for my study.But there are some trouble for me about the data preprocessing.Could you tell me some detail about how you do the data preprocessing?

There is a little mistake in computer_error.py

In https://github.com/guohengkai/region-ensemble-network/blob/master/evaluation/compute_error.py line 38

results[idx] = np.where(err_flat <= th)[0].shape[0] * 1.0 / err_flat.shape[0]

should be :
err_flat = errs # errs.shape=(8252,14)
results[idx] = np.where(np.max(err_flat, axis=1 <= th)[0].shape[0] * 1.0 / err_flat.shape[0]

solver.prototxt

Dear hengkai,
Can you share the solver.prototxt file that used to train the networks? I want to run the architecture described in the paper.
regards,
weiguo

Visualization part lack argumentparser

Hello,
Thank you for sharing! There may be needed more argumentparsers in the visualization part,e,g,dataset_model.

$ python evaluation/show_result.py icvl --in_file=results/icvl_ren_4x6x6.txt

regards,
weiguo

hand_model.detect_image() error

Hi,

Thank you for your nice code! I want to reproduce the results, Since the pyrs.start() doesn't work, then I am using pysr.service instead. There is some following errors

Traceback (most recent call last):
File "demo/show_cv2.py", line 75, in
results = hand_model.detect_image(d)
File "/home/alex/PhD/PoseEstimation/REN/region-ensemble-network-master/evaluation/hand_model.py", line 42, in detect_image
return self.detect_images([img])[0,...]
TypeError: 'NoneType' object has no attribute 'getitem'

Can you give me some advice?

Regards,
weiguo

About the loss function

hi:
excellent job！！
I want to retrain the model , is it convenient for you to tell me the loss function.
I did not see the info about loss function neither in the paper nor in the model folder(only deploy prototxt)
thank you !

trianing code about ITOP dataset

hi, thx for you work. Can you release the training code about ITOP dataset? I'm dong some research on this
dataset and looking forward to it.

label normalization

thx a lot for your paper!
Here comes the question,how you normalized the joints according to the cropping?
thx again and hope for your reply.

nyu数据集中joint_uvd中深度值与深度图中对应位置深度的值不匹配？

您好，我发现从joint_uvd中读出的关节点的深度值与对应的深度图中相对的关节点位置处的深度值是不一样，请问这是怎么回事？

parse model error.

感谢分享~
我在执行： python evaluation/run_model.py icvl ren_4x6x6 your/path/to/output/file your/path/to/ICVL/images/test 的时候，
读入模型报了下面的错误：
Can't parse message of type "caffe.NetParameter" because it is missing required fields: layer[53].psroi_pooling_param.output_dim, layer[53].psroi_pooling_param.group_size

nyu数据集下载总是中断，请问除了官方的数据源还有没有其他下载方式或者国内的镜像？十分感谢！！

Annotation error on NYU dataset

Hi, thank you for your code. And it works well on ICVL dataset. However, I found there is some error on NYU dataset. Hand joint position is correct, but the line is error with the start point. Can you give the more information on annotation of 14 hand pose joints of the NYU dataset which you used to train?

def get_sketch_setting(dataset):
    if dataset == 'icvl':
        return [(0, 1), (1, 2), (2, 3), (0, 4), (4, 5), (5, 6),
                (0, 7), (7, 8), (8, 9), (0, 10), (10, 11), (11, 12),
                (0, 13), (13, 14), (14, 15)]
    elif dataset == 'nyu':
        return [(0, 1), (0, 2), (0, 5), (3, 4), (4, 5), (0, 7), (6, 7),
                (0, 9), (8, 9), (0, 11), (10, 11), (0, 13), (12, 13)]
    elif dataset == 'msra':
        return [(0, 1), (1, 2), (2, 3), (3, 4), (0, 5), (5, 6), (6, 7), (7, 8),
                (0, 9), (9, 10), (10, 11), (11, 12), (0, 13), (13, 14), (14, 15), (15, 16),
                (0, 17), (17, 18), (18, 19), (19, 20)]