| | |
| | | free_layer(net.layers[i]); |
| | | } |
| | | free(net.layers); |
| | | |
| | | free(net.scales); |
| | | free(net.steps); |
| | | free(net.seen); |
| | | |
| | | #ifdef GPU |
| | | if (gpu_index >= 0) cuda_free(net.workspace); |
| | | else free(net.workspace); |
| | |
| | | int f; |
| | | for (f = 0; f < l->n; ++f) |
| | | { |
| | | l->biases[f] = l->biases[f] - l->scales[f] * l->rolling_mean[f] / (sqrtf(l->rolling_variance[f]) + .000001f); |
| | | l->biases[f] = l->biases[f] - (double)l->scales[f] * l->rolling_mean[f] / (sqrt((double)l->rolling_variance[f]) + .000001f); |
| | | |
| | | const size_t filter_size = l->size*l->size*l->c; |
| | | int i; |
| | | for (i = 0; i < filter_size; ++i) { |
| | | int w_index = f*filter_size + i; |
| | | |
| | | l->weights[w_index] = l->weights[w_index] * l->scales[f] / (sqrtf(l->rolling_variance[f]) + .000001f); |
| | | l->weights[w_index] = (double)l->weights[w_index] * l->scales[f] / (sqrt((double)l->rolling_variance[f]) + .000001f); |
| | | } |
| | | } |
| | | |