DCGans

DCGans

1. Load_SVHN(street view housing numbers) { matlab files}

2. scale(x,feature_range=(-1,1)) { //convert input images in the range of -1 to 1 }

3. batches(batch_size){}

4. model_Inputs(real_dim,z_dim) { return input_real ,input_z}

5. generator(z,out_dim,reuse=False,alpha=0.2,training=True) { fully_connected_layer reshape-->batch_normal-->leaky_relu(4*4*512) conv_trans-->batch_normal-->leaky_relu(8*8*256) conv_trans-->batch_normal-->leaky_relu(16*16*128) logits=conv_trans(32*32*3) out=tanh(logits) return out}

6. discriminator(z,reuse=False,alpha=0.2) { #input(32*32*3) conv-->leaky_relu(16*16*32) conv_trans-->batch_normal-->leaky_relu(8*8*128) conv_trans-->batch_normal-->leaky_relu(4*4*256) reshape(to_flatten) logits=fully_connected_layer out=sigmoid(logits) returnout,logits }

7. Model_Loss(input_real, input_z, output_dim, alpha=0.2) { d_loss,g_loss}

8. Model_optimizers(d_loss, g_loss, learning_rate, beta1) { d_tran_optim,g_train_optim}