| 18 | self._comp_mask = self.def_comp_mask() |
| 19 | |
| 20 | def def_comp_mask(self): |
| 21 | BS = self.BS |
| 22 | print('COMPILING') |
| 23 | t = time() |
| 24 | m = T.tensor4() |
| 25 | bf_w = np.ones((1, 1, 2 * BS, 2 * BS)) |
| 26 | bf = sharedX(floatX(bf_w)) |
| 27 | m_b = dnn_conv(m, bf, subsample=(BS, BS), border_mode=(BS / 2, BS / 2)) |
| 28 | _comp_mask = theano.function(inputs=[m], outputs=m_b) |
| 29 | print('%.2f seconds to compile [compMask] functions' % (time() - t)) |
| 30 | return _comp_mask |
| 31 | |
| 32 | def comp_mask(self, masks): |
| 33 | masks = np.asarray(self._comp_mask(masks)) |