A Large-Scale Dataset for Robotic Grasp Detection

Example of images from Jacquard

Request access

A small sample of 10 objects can be downloaded from the Database page. To acquire access to the whole dataset:

After your request is verified and approved by a website administrator, you will receive a notification by email. Please note, that any requests from free email addresses (hotmail, yahoo, gmail, etc.) will be refused.

With the Jacquard dataset, we propose a new criterion based on simulation, subsequently called simulated grasp trial-based criterion (SGT). Specifically, when a new grasp should be evaluated, the corresponding scene is rebuilt in the simulation environment and the grasp is performed by the simulated robot, in the same conditions as during the generation of the annotations. If the outcome of the simulated grasp is a success, i.e., the object is successfully lifted and moved away by the simulated robot using the predicted grasp location, the prediction is then considered as a good grasp.

Along with the access to the whole dataset, registered users can use an interface to submit their predictions to our simulator and get the result of the Simulated Grasp Trial-based criterion introduced in the paper.


If you use Jacquard Dataset in your research, please cite the following paper:


The Jacquard Dataset has been created by a french team of researchers:

We want to thank Matthieu Grard and Romain Brégier for their help. Of course, we also would like to thank the creators of Shapenet who shared their work.