FruitBin - version1


FruitBin contains more than 1M images and 40M instance-level 6D pose annotations over both symmetric and asymmetric fruits with or without texture. Rich annotations and metadata (including 6D pose, segmentation mask, point cloud, 2D and 3D bounding boxes, occlusion rate) allow the tuning of the proposed dataset for benchmarking the robustness of object instance segmentation and 6D pose estimation models (with respect to variations in terms of  lighting, texture, occlusion, camera pose and scenes). We further propose three scenarios presenting significant challenges of 6D pose estimation models: new scene generalization; new camera viewpoint generalization; and occlusion robustness. We show the results of these three scenarios for two 6D pose estimation baselines making use of RGB or RGBD images. To the best of our knowledge, FruitBin is the first  dataset for the challenging task of fruit bin picking and the biggest large-scale dataset for 6D pose estimation with the most comprehensive challenges, tunable over scenes, camera poses and occlusions.

License : CC BY-NC-SA

Download instructions

The data is initially divided into seven subsets, each corresponding to a different point of view. For each subset, the data can be accessed in two ways:

  • Raw Data: This can be found in the “Data” folder, which contains zip files organized by data type (Bbox, RGB images, depth, segmentations, occlusion, etc.).

  • Benchmark Data: This contains benchmark scenarios and can be found in eight zip files, each corresponding to a specific fruit.

Each fruit folder contains a set of features such as Bbox, Depth, FPS, Instance_Mask, Labels, Meta, Models, Pose_transformed, and RGB_resized. Some folders may have names ending in 'Gen' or 'Resized' due to image size reduction for training purposes. In addition to that, the 'splitting' folder contains text files that list the images related to the 8 scenarios, along with the train/eval/test splitting. These files are named [split]_[fruit_name]_[scenario_type]_[lower_bound_visibility]. For more precise splitting over occlusion, additional folders named 'splitting_occ_01_cameras/worlds' provide additional splitting files as described in our 'FruitBin' paper.

Download from
Publication date
Guillaume Duret, Mahmoud Ali, Nicolas Cazin, Alexandre Chapin, Florence Zara, Emmanuel Dellandrea, Jan Peters, Liming Chen
Dataset size
Raw data ~7x250 Go, Benchmarks ~7x20 Go