search

skumra / robotic-grasping

Antipodal Robotic Grasping using GR-ConvNet. IROS 2020.
Other
455 stars 119 forks source link

Availability of pretrained models #5

Closed sallogy closed 4 years ago

sallogy commented 4 years ago

Hi Sulabh,

great work, congratulations! Can you please make available the prettained model cornell_rgbd_iou_0.96 which is referred in run_grasp_generator.py.

Thanks, György Salló

skumra commented 4 years ago

Hi György,

Thank you for the interest in our research. We will be uploading a bunch of good pre-trained models once the review process of our paper is complete.

You can also use the train_network.py to train a new model yourself.

Thanks, Sulabh

sallogy commented 4 years ago

Hi Sulabh,

thanks for the fast response. What do you expect, when the review process will be finished? I will try to train the model myself - can you tell me what GPU resources are needed, and approximately how long it will take?

Thanks, György

skumra commented 4 years ago

We expect the review process to be completed by June 30, 2020. Given the light weight nature of GR-ConvNet, you can train the model within a few hours using a decent Nvidia GPU. You can also train on a CPU but it may take up to a day to train. Also, training the model using only depth data will be a lot faster.

sallogy commented 4 years ago

Hi Sulabh,

thank you for the info.