search

anatolix / keras_Realtime_Multi-Person_Pose_Estimation

Keras version of Realtime Multi-Person Pose Estimation project
Other
118 stars 47 forks source link

Inspect dataset not working. #10

Open piperod opened 6 years ago

piperod commented 6 years ago

Im trying to visualize the data of the generated dataset h5. First I generated a new h5 following the instructions for new generated dataset. Then I was trying to visualize the augmentation using the recomendation from the readme.

  1. I try using the notebook in the section "Verify data generator":

    from ds_iterator import DataIterator
di = DataIterator("../dataset/val_dataset_new.h5", data_shape=(3, 368, 368),
              mask_shape=(1, 46, 46),
              label_shape=(57, 46, 46),
              vec_num=38, heat_num=19, batch_size=10, shuffle=True)
x,y = di.next()
x,y = di.next()
x,y = di.next()

    However it raises an error No module named ds_iterator

  2. I also tried using "./rmpe_servert_tester.py raw save" But it didn't process anything nor show any error. Before this I modified the function main() in rmpe_server.py with 

    val =    Server(["../dataset/val_dataset_new.h5"],        [RmpeCocoConfig], 5556, "Val",   shuffle=False, augment=False)"

Maybe I didn't understand all the instructions. Is it a different part of the notebook the one I should use? or is the command for save the images to disk needs me to create a folder or take any further step previously?

anatolix commented 6 years ago
  1. I may have changes name of files and haven't updated notebook. Sorry it is ds_generators.py now. Will be grateful if you fix and make pull request.
  2. ./rmpe_servert_tester.py raw save waiting to connect rmpe_server.py which should be run on 5556 port. It could be mine or old c++ server.
  3. To be honest I've moved to another branch with development which called new-generation and mostly use notebooks from there https://github.com/anatolix/keras_Realtime_Multi-Person_Pose_Estimation/blob/new-generation/testing/inspect_nn_input_output.ipynb Contrary to inspect-dataset they inspect not content of hdf5, but data we feed to neural network and get from neural network.