This is the repository for releasing trained CNN models for gravity prediction of the paper, "Force from Motion: Decoding Physical Sensation from a First Person Video" in CVPR 2016.
For more information, please visit our project webpage: http://www.seas.upenn.edu/~hypar/ffm.html
We use Caffe deep learning framework: http://caffe.berkeleyvision.org/
The CNN models are fine-tuned from ImageNet-pretrained AlexNet with the following input/output modifications:
-
Image resolution is (180, 320), instead of (227, 227).
-
Output of the network is 61 dimensions. It predicts the probability of projected gravity angle discretized by 1 degree between -30 and 30 with the 31th dimension as 0 degree.
Please check the sample prototxt file.
There are 3 models fine-tuned for different scenarios: biking (taxco), skiing, and jet-skiing.
Due to the size limitation of Github, please download the trained models from: https://upenn.box.com/v/gravity