A Deep Learning Framework for Character Motion Synthesis and Editing




  • Paper topic: Animation and Simulation
  • Software type: Code
  • Able to run a replicability test: False
  • Replicability score: 0
  • Software language: Python
  • License: Custom
  • Build mechanism:
  • Dependencies: theano,numpy,scipy,six,matplotlib
  • Documentation score {0,1,2}: 1
  • Reviewer: Nicolas Bonneel <nicolas.bonneel@liris.cnrs.fr>
  • Time spent for the test (build->first run, timeout at 100min): 100min

Source code information


I could not run the code due to missing library versions. Given the code syntax, I assumed the code was Python 3, and given the date of the paper, I tried Python 3.5. I initially attempted with theano 0.8.2 (which installs  numpy-1.18.5 scipy-1.4.1 six-1.15.0, though I had to downgrade scipy to 1.3 to avoid some 'fft' library error), but I ended up with numpy errors ('Could not import inplace_increment' with Theano asking me to update to NumPy 1.8 though I already had 1.18 (following internet's advice, I installed libpython but this did not solve the problem). Given that Theano 0.8.2 was released after the Siggraph 2016 deadline, I also tried with Theano 0.7 which required me to install a GCC on windows (I used https://jmeubank.github.io/tdm-gcc/download/ though I had to edit their cmath file, line 1121 to replace 'using ::hypot;' to 'using ::_hypot;') but I ran into further errors ('ImportError: DLL load failed: A dynamic link library (DLL) initialization routine failed.' when calling 'index = rng.randint(X.shape[0].eval())' in demo_stepped ; or 'ERROR (theano.gof.opt): Optimization failure due to: constant_folding' in demo_style_transfer and demo_basis).
 Trying with Theano 0.6 resulted in other errors (KeyError: 'blas' and 'blas.ldflags' in Python's configparser.py) which resulted me to install 2018 and 2019 versions of mkl following internet's advice, but without further success (2017 was not available on my pip).
 The data file should be downloaded and extracted such that the folder motionsynth_data/data should be at the same level as the other directories motion, nn and synth. 

 Since there is a combinatorial number of library versions and dependencies (perhaps Python 3.4 was needed?), it would be extremely useful if the authors could provide the versions of the libraries used in their project.

Misc. comments

An unofficial partial reimplementation is also available at:

If you want to contribute with another review, please follow these instructions.

Please consider to cut/paste/edit the raw JSON data attached to this paper.