pathak22/zeroshot-imitation

[ICLR 2018] TensorFlow code for zero-shot visual imitation by self-supervised exploration

zeroshot-imitation's Issues

Can you provide the code of the multi-step GSP which mentioned in the paper?

I meet two problems while running the code.

1. The network architecture in train.py only supports the inverse model(single-step output), and there is no code for multi-step output. Can you provide the code of Multi-step GSP(multi-step output model)?
2. The format of datasets download from google drive does not match the rope_data.py, such as the download datasets do not have the /image_before, /image_after, /poke. Can you provide the data processing script or the datasets that match the code?

Waiting for your reply! @pathak22

Is there a data preprocessing script for the rope dataset?

Firstly, thank you for your great work!

The format of provided dataset google drive link is not completely fit to the training code / dataloader. For example, there is no image_before/image_after splits in the rope dataset which is used for the rope_data.py. Could you please provide the data preprocessing tools for the training code?

Thank you!

print() is a function in Python 3

flake8 testing of https://github.com/pathak22/zeroshot-imitation on Python 3.6.3

$ flake8 . --count --select=E901,E999,F821,F822,F823 --show-source --statistics

./data/database.py:76:27: E999 SyntaxError: invalid syntax
                print "key", cursor.set_key(label)
                          ^
./data/rope_data.py:52:32: E999 SyntaxError: invalid syntax
            print "loading from", r, "size", db_pokes.i,
                               ^
./data/window.py:24:25: F821 undefined name 'ind'
    return ret, action, ind
                        ^
./nets/alexnet_geurzhoy.py:100:21: E999 SyntaxError: invalid syntax
        print "REUSE", reuse
                    ^
3     E999 SyntaxError: invalid syntax
1     F821 undefined name 'ind'
4

Google drive link does not contain img_mean.npy

Only rope.zip is available from the google drive link in the README.

Can you update the README to provide the link to img_mean.npy?

Is there a demo?

How can I use the trained model to make the turtlebot to navigate to the goal image like the video? May you offer a demo?Thanks.

can you please explain about the group parameter in def conv

where is the goal recognizer part?

do you have the data set for the navigation task?