eldar / pose-tensorflow

Human Pose estimation with TensorFlow framework
GNU Lesser General Public License v3.0
1.14k stars 384 forks source link

dataset-test.mat is missing #11

Open paragjai opened 7 years ago

paragjai commented 7 years ago

Where is dataset-test.mat is present which _loaddataset method of PoseDataset class of _posedataset.py is calling?

cfg.dataset is equal to dataset-test.mat

Traceback (most recent call last): File "test.py", line 75, in test_net(not args.novis, args.cache) File "test.py", line 20, in test_net dataset = create_dataset(cfg) File "/home/paragjain/Desktop/pose-tensorflow/dataset/factory.py", line 22, in create data = PoseDataset(cfg) File "/home/paragjain/Desktop/pose-tensorflow/dataset/pose_dataset.py", line 53, in init self.data = self.load_dataset() File "/home/paragjain/Desktop/pose-tensorflow/dataset/pose_dataset.py", line 67, in load_dataset mlab = sio.loadmat(file_name) File "/home/paragjain/anaconda3/envs/tensorflow2/lib/python2.7/site-packages/scipy/io/matlab/mio.py", line 135, in loadmat MR = mat_reader_factory(file_name, appendmat, **kwargs) File "/home/paragjain/anaconda3/envs/tensorflow2/lib/python2.7/site-packages/scipy/io/matlab/mio.py", line 58, in mat_reader_factory byte_stream = _open_file(file_name, appendmat) File "/home/paragjain/anaconda3/envs/tensorflow2/lib/python2.7/site-packages/scipy/io/matlab/mio.py", line 23, in _open_file return open(file_like, 'rb') IOError: [Errno 2] No such file or directory: 'dataset-test.mat'

YYfangzi commented 7 years ago

Have the same problem

eldar commented 7 years ago

Hi guys. I haven't uploaded the code, that generates the test set yet. Is it what you are trying to do? Because otherwise, just for predictions you don't need the test set - you can just use the demo code as template.

YYfangzi commented 7 years ago

@eldar Thank you for very much! The MPII dataset has about 25k images.I want to know the training code you provided is on these whole 25k images and how about its test dataset? The model is acquired from training on the train dataset and then the model is used on the test dataset to get evaluation results such as mAP. The demo code can provide visual results just on a single image but for evaluation, the model got from trainning may be used on the test set. We want to train on our own dataset and then evaluation on our test set ,so we just have those problems.Thank you for your attention!

Mourula commented 7 years ago

@eldar How should we preprocess the image(s) that we feed into the demo code? I guess you have preprocessed (the scaling and cropping) the training and testing images based on the annotations' spots (head size etc.)?

gsoykan commented 3 years ago

@eldar I am also suffering from the same issue. Would you please upload the code that generates the test set? I am trying to replicate your work with Julia Lang by using Knet platform. I would really appreciate if you do upload the code, because I have already developed and trained my neural network. However, I can not test it.

Ai-econ commented 8 months ago

@eldar, Thanks for splendid work for pose-estimation. Would you upload the code that generates the test set?