This video is a presentation developed as for the final exam of the "Computational approaches to Physical and Virtual Crowd Phenomena" course by Prof. Giuseppe Vizzari.  We used Autodesk 3Ds Max with the Populate Toolset, the CAFFE Neural Network framework and the CRFASRNN approach to detect the people shapes inside a simulated photorealistic environment.

In this second video the results of the CRFASRNN model is shown. We apply the model without any further training. The results suggest a further training is needed for such environment.

Download the dataset 

• The
  The Walking Dataset (685 downloads )
  with Training Masks
  Walk01 - Fullsize (654 downloads )
  and 
  Walk01 - Resized (776 downloads )
• The 
  3ds Max Scenes (711 downloads )
• All videos used in the above presentation (762 downloads )

First dataset training mask

References
http://caffe.berkeleyvision.org [official website of the NN framework]
http://www.robots.ox.ac.uk/~szheng/CRFasRNN.html [the CRF as RNN approach website]
https://github.com/martinkersner/train-CRF-RNN [scripts for caffé training purposes]
https://knowledge.autodesk.com/support/3ds-max [educational links provided]