[Project] [Paper] [Dataset]

Checkout the github repository to download the YCB-Affordance dataset. This contains the 3D models of objects from the YCB benchmark, the videos from the YCB-Video Dataset, and the human hand grasps from the YCB-Affordance dataset.

• conda create -n ganhand python=3.6
• conda activate ganhand
• Python requirements: Run pip install -r requirements.txt.
• MANO layer: Follow instructions from the MANO layer project in here.

• Download the YCB-Affordance Dataset from this repository. We use the YCB-Affordance Dataset, an extension of the YCB-Video dataset, to train and test the model. Follow the instructions of the repository to download the dataset. Link this project to the folder where you keep the dataset using the data_dir argument, when training/testing.
• We use https://github.com/cvlab-epfl/segmentation-driven-pose to estimate the pose of the objects in the YCB-Video dataset . You can download the predictions for all train and test samples from the YCB dataset in https://drive.google.com/file/d/17WEN8vhel6Ico-lGDtrcxu7XFemVi4qN/view, and move them in the YCB Affordance dataset folder

GanHand takes a single RGB image of one or several objects and predicts how a human would grasp these objects naturally. Our architecture consists of three stages. First, the objects' shapes and locations are estimated in the scene using an object 6D pose estimator or a reconstruction network (red). The predicted shape is then projected onto the image plane to obtaina segmentation mask that is concatenated with the input image and fed to the second sub-network for grasp prediction (blue). Finally, werefine the hand parameters and obtain hand final shapes and poses using a differentiable parametric model MANO (yellow). The model is trained using adversarial, interpenetration, classification and optimization losses, indicated in bold.

Videos like those in the teaser of the paper can be obtained running the following command. The pretrained model can be downloaded from this link and placed under a folder named checkpoints. So the main folder should contain the model checkpoints in /checkpoints/ganhand_pretrained/

python test.py --dataset_mode ycb_affordances_complete_scene --name ganhand_pretrained --load_epoch 13

• In this paper, we make extensive use of the differentiable MANO model
• We follow the deep learning model framework from IRI-DL

If this dataset is useful in your research, please cite:

@inproceedings{corona2020ganhand,
  title={Ganhand: Predicting human grasp affordances in multi-object scenes},
  author={Corona, Enric and Pumarola, Albert and Alenya, Guillem and Moreno-Noguer, Francesc and Rogez, Gr{\'e}gory},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={5031--5041},
  year={2020}
}

The YCB-Affordance dataset is released only for research purposes