GibsonEnv icon indicating copy to clipboard operation
GibsonEnv copied to clipboard

Published 20 hours ago • verified publisher icon StanfordVL

Gibson Goggle

Open jhpenger opened this issue 6 years ago • 3 comments

Can you guys provide more instructions on how to use Gibson's Goggle? I am trying to convert RGB images to Gibson style to train an image classifier for identifying chairs.

I am running Gibson headless in a docker container. Thanks,

jhpenger avatar Dec 12 '18 00:12 jhpenger

Same question here. I'm trying to retrain Goggles on a different crop dimension of the original dataset. But I'm having trouble finding the correct root directory to load the data, as done in lines 100 and 101 of train_fillers.py. I've downloaded the data using .download.sh, but don't know which directory under the data folders I should run PairDataset on. Any help would be greatly appreciated!

ianhuang0630 avatar Dec 23 '18 23:12 ianhuang0630

@jhpenger One example of using gibson goggle can be found in https://github.com/StanfordVL/GibsonEnv/blob/master/examples/ros/gibson-ros/goggle.py. It requires a kinect camera and ROS, as the goggle accepts RGBD images and adapts into to gibson style images.

fxia22 avatar Dec 27 '18 06:12 fxia22

@fxia22 Can you provide some commentation for that file? I'm not using video feed from a camera, just a bunch of chair images to train an image classifier. Or can you provide a very easy example of converting normal jpg/png to gibson style? Thanks,

jhpenger avatar Dec 29 '18 09:12 jhpenger