Pretrained model

[sirykd]

Hi! It would be great if you could share a pretrained model, which people would be able to run and get the results without spending 2+ days on training.

[wbenbihi]

Hi, You can download a pretrained model here Tell me If you have any trouble, some tools should be available soon for prediction and visualization.

[bhack]

Is there not too much duplicated/overlapping code now between hourglass_tiny and hourglass_tiny_testing?

[wbenbihi]

hourglass_tiny_testing is a backup file. You cannot use the new hourglass_tiny with the pretrained model and it is not compatible with the predictClass.py. I am making a cleaner code. Should be available at the end of the week.

[bhack]

Yes I'have seen this.. but was just a feedback about how the code was organized between this two files.

[bhack]

There are missing Yolo python imports.. Another alternative is to use one of the models zoo available in TF new Object detection API

[bhack]

Now yolo code is imported in the repository but you are mixing your code under MIT with unkown yolo code license. This version is under the same of your code but I've not test it. Probably could better to separate in two files the code that requires a detector form the code that doesn't require one (I.e. single person methods etc.)

[o0t1ng0o]

Hi~ @wbenbihi There are only two files about the pretrained model. (hourglass_bn_100.data-00000-of-00001 hourglass_bn_100.index)

But, when i restore the your pretrained model, i need the file suffixed by ".meta".

Could you please offer this file?

[cardcaster]

Hi, I'm new to tensorflow and I'm trying to use the pretrained model called Regular Hourglass (80 Convolution layer) (ref/refined_tiny). I read the readme.txt and as instructed I edited hourglass_tiny.py and changed def restore(self, load = None):

to

def restore(self, load = 'saver_directory'):

My config.cfg file looks like this

[DataSetHG]
training_txt_file: 'dataset.txt'
img_directory: '/Users/cardcaster/Downloads/hourglasstensorlfow-master/images/'
img_size: 256
hm_size: 64
num_joints: 16
remove_joints: None
joint_list = ['r_anckle', 'r_knee', 'r_hip', 'l_hip', 'l_knee', 'l_anckle', 'pelvis', 'thorax', 'neck', 'head', 'r_wrist', 'r_elbow', 'r_shoulder', 'l_shoulder', 'l_elbow', 'l_wrist']`
[Network]
name: 'hg_refined'
nFeats: 256
nStacks: 4
nModules: 1
tiny: False
nLow: 4
dropout_rate: 0.2
mcam: False
[Train]
batch_size: 4
nEpochs: 200
epoch_size: 1000
learning_rate: 0.00025
learning_rate_decay: 0.96
decay_step: 2000
weighted_loss: False
[Validation]
valid_iteration: 10
[Saver]
log_dir_train: 'F:/Cours/DHPE/DHPE/hourglass_tiny/'
log_dir_test: 'F:/Cours/DHPE/DHPE/hourglass_tiny/'
saver_step: 500
saver_directory: '/Users/cardcaster/Downloads/hourglasstensorlfow-master/pretrained-model/'

However, as mentioned in the readme.txt I should use the same config file but the pretrained model comes with its own config.cfg file. Which one should I be using?

Thank you.

[Ziang-Wang]

嗨， 您可以在这里下载经过训练的模型。 告诉我，如果有任何麻烦，应尽快提供一些工具来进行预测和可视化。

hi, the link does not exist. That would be great if you could share pretrained models again. thank you so much.

[ICEAIyjp]

It would be great if you could share a pretrained model again, please，the old website doesn't work.