Sample Weight Models for Predictive Coding Deep Neural Network (Front. Psychol., 15 March 2018), and a Bug-fixed Version
Sample weight model used in Illusory Motion Reproduced by Deep Neural Networks Trained for Prediction (Front. Psychol., 15 March 2018; https://doi.org/10.3389/fpsyg.2018.00345).This model was generated with predictive coding deep neural network using 500K FPSI video images.
[Important remarks] About 29v_300K_i10K.model
The video used in the paper has a bug where it plays the same two frames in a row (which may not be a bug for artificial perception...). The model 29v_300K_i10K.model was trained on the video with the bug fixed. The GPU parameters and initial values were fixed for 29v_300K_i10K.model training.
The videos used for the training were shared below.https://drive.google.com/file/d/1i8Iv4Gd9zwVyvgXTtVb4RIHeex4D5tXr/view?usp=sharing
510k_vga.model = a 640x480 high resolution model