Download and resources


This dataset is released under the Creative Commons Attribution 4.0 International License.


When using the IntPhys dataset for publications, please cite our paper on arxiv


Click on the links below to download the intphys datasets.

  • Train dataset includes 15k scenes with full metadata.
  • Test datasets include 1080 scenes for each block without metadata.
  • Development dataset includes 30 scenes for each block with partial metadata (physical plausibility of the scene is given).
  • The entire dataset has a size of about 250G.
File Description Size MD5 sum Challenge starting kit 180K cf699d546892d206c51ceff352a2534c
dev.tar.gz Development data (all blocks) 3G 6a72715007f2f3b0e9546bfa8d8fc39b
test.O1.tar.gz Test data (block O1) 36G cc998a7a1b14bba988bd1ef31ae6cd6b
test.O2.tar.gz Test data (block O2) 36G d6d8475bfd06298e938b8925ad326e7e
test.O3.tar.gz Test data (block O3) 36G 318445e082357e1256eff63a96348c8b
train.1.tar.gz Train data (1/4) 35G 3428bcf398441a489111528634530212
train.2.tar.gz Train data (2/4) 35G 47dc1c3578e079584f4e3b5b8ab8304b
train.3.tar.gz Train data (3/4) 35G 389782958ad058f7a7e1db2b07748259
train.4.tar.gz Train data (4/4) 35G 7d345eaa3a33d2abc4a75442eefb277a

Additional resources

   author    = {Ronan Riochet and
                Mario Ynocente Castro and
                Mathieu Bernard and
                Adam Lerer and
                Rob Fergus and
                V{\'{e}}ronique Izard and
                Emmanuel Dupoux},
   title     = {IntPhys: {A} Framework and Benchmark for Visual Intuitive Physics
   journal   = {CoRR},
   volume    = {abs/1803.07616},
   year      = {2018},
   url       = {},
   archivePrefix = {arXiv},
   eprint    = {1803.07616},
   timestamp = {Mon, 13 Aug 2018 16:47:10 +0200},
   biburl    = {},
   bibsource = {dblp computer science bibliography,}