TransProteus, Predicting 3D shapes, masks, and properties of materials, liquids, and objects inside transparent containers from images

We present TransProteus, a dataset, for predicting the 3D structure and properties of materials, liquids, and objects inside transparent vessels from a single image without prior knowledge of the image source and camera parameters. Manipulating materials in transparent containers is essential in many fields and depends heavily on vision. This work supplies a new procedurally generated dataset consisting of 50k images of liquids and solid objects inside transparent containers. The image annotations include 3D models and material properties (color/transparency/roughness...) for the vessel and its content. The synthetic (CGI) part of the dataset was procedurally generated using 13k different objects, 500 different environments (HDRI), and 1450 material textures (PBR) combined with simulated liquids and procedurally generated vessels. In addition, we supply 104 real-world images of objects inside transparent vessels with depth maps of both the vessel and its content.

Note that there are two files here:

Transproteus_SimulatedLiquids2_New_No_Shift.7z

and

TranProteus2.7z , contain subset of the virtual CGI data set.

https://zenodo.org/api/files/12b013ca-36be-4156-afd4-c93b5fa22093/Tansproteus_SimulatedLiquids2_New_No_Shift.7z

TransProteus_RealSense_RealPhotos.7z : Contain real-world photos scanned with real sense with depth map of both the vessel and its content

See ReadMe file in side the downloaded files for more details

The full dataset (>100gb) can be found here:

https://e.pcloud.link/publink/show?code=kZfx55Zx1GOrl4aUwXDrifAHUPSt7QUAIfV

https://icedrive.net/1/6cZbP5dkNG

See:  https://arxiv.org/pdf/2109.07577.pdf for more details

**This dataset is complementary to LabPics dataset with 8k real images of materials in vessels in chemistry labs, medical labs, and other settings. The LabPics dataset can be downloaded from here:

https://zenodo.org/record/4736111#.YVOAx3tE1H4

************************************************************************************

Transproteus_SimulatedLiquids2_New_No_Shift.7z and TranProteus2.7z

The two folders contain relatively similar data styles.
The data in No_Shift contain images that were generated with no camera shift in the camera paramters. If you try to predict 3d model from an image as a depth map, this is easier to use (Otherwise, you need to adapt the image using the shift). For all other purposes, both folders are the same, and you can use either or both. In addition, a  real image dataset for testing is given in the RealSense file.