saveImage(output_list, node_shape[2], 'cnn2_before_train') cnn2.pre_train() output_list = cnn2.output() saveImage(output_list, node_shape[2], 'cnn2_after_train') rbm_size_list = (680, 340, 170, 85, 42, 21, 10, 3) # def __init__(self, W, input, data_size,input_size, output_size, isDropout): rbm1 = RBM(None, cnn2.output(), file_num, rbm_size_list[0], rbm_size_list[1], False) for i in xrange(pre_train_epoch): print 'rbm1 pre_train:' + str(i) rbm1.contrast_divergence() reinput = rbm1.reconstruct_from_input(rbm1.input) saveImage(reinput, node_shape[2], 'rbm1_after_train') saveW(rbm1.getW(), 'rbm1_after_train') rbm2 = RBM(None, rbm1.output(), file_num, rbm_size_list[1], rbm_size_list[2], False) for i in xrange(pre_train_epoch): print 'rbm2 pre_train:' + str(i) rbm2.contrast_divergence() reinput = rbm2.reconstruct_from_input(rbm2.input) reinput = rbm1.reconstruct_from_output(reinput) saveImage(reinput, node_shape[2], 'rbm2_after_train') saveW(rbm2.getW(), 'rbm2_after_train') rbm3 = RBM(None, rbm2.output(), file_num, rbm_size_list[2], rbm_size_list[3], False) for i in xrange(pre_train_epoch): print 'rbm3 pre_train:' + str(i) rbm3.contrast_divergence() reinput = rbm3.reconstruct_from_input(rbm3.input)
result_path = 'data/kouryu_room/cnn2_after_training' result_data = load_result_image(result_path, file_num, isRGB) # result_path = 'data/4position_rumba/image7000/rbm1_train3434' # result_W = loadW(result_path) makeFolder() # def __init__(self, W, input, data_size,input_size, output_size, isDropout): rbm1 = RBM(None, result_data, file_num, rbm_size_list[0], rbm_size_list[1]) for i in xrange(pre_train_epoch): print 'rbm1 pre_train:' + str(i) rbm1.contrast_divergence(i) reinput = rbm1.reconstruct_from_input(rbm1.input) saveImage(reinput, node_shape[2], 'rbm1_after_train') saveW(rbm1.getW(), 'rbm1_after_train') rbm2 = RBM(None, rbm1.output(), file_num, rbm_size_list[1], rbm_size_list[2]) for i in xrange(pre_train_epoch): print 'rbm2 pre_train:' + str(i) rbm2.contrast_divergence(i) reinput = rbm2.reconstruct_from_input(rbm2.input) reinput = rbm1.reconstruct_from_output(reinput) saveImage(reinput, node_shape[2], 'rbm2_after_train') saveW(rbm2.getW(), 'rbm2_after_train') rbm3 = RBM(None, rbm2.output(), file_num, rbm_size_list[2], rbm_size_list[3]) for i in xrange(pre_train_epoch): print 'rbm3 pre_train:' + str(i) rbm3.contrast_divergence(i) reinput = rbm3.reconstruct_from_input(rbm3.input)