def on_epoch_end(self, epoch, logs=None): if logs: for key in logs.keys(): if epoch == 0: self.history[key] = [] self.history[key].append(logs[key]) if epoch % 5 == 0: plot_history_for_callback(model_folder + '/train_history.png', self.history) save_history_for_callback(model_folder, self.history) img_vols, gt_vols, pr_vols = [], [], [] for i in range(0, len(valid_dataset), int(len(valid_dataset) / 64)): img_vols.append(np.load( valid_dataloader.dataset.images_fps[i])) gt_vols.append(valid_dataloader[i][1]) pr_vols.append(self.model.predict(valid_dataloader[i])) img_vols = np.stack(img_vols, axis=0) gt_vols = np.concatenate(gt_vols, axis=0) pr_vols = np.concatenate(pr_vols, axis=0) save_images(model_folder + '/epoch-{}-img.png'.format(epoch), np.uint8(img_vols)) save_images(model_folder + '/epoch-{}-gt.png'.format(epoch), gt_vols / args.scale * 255) save_images(model_folder + '/epoch-{}-pr.png'.format(epoch), pr_vols / args.scale * 255)
def on_epoch_end(self, epoch, logs=None): if logs: for key in logs.keys(): if epoch == 0: self.history[key] = [] self.history[key].append(logs[key]) if epoch % 5 == 0: plot_history_for_callback(model_folder + '/train_history.png', self.history) save_history_for_callback(model_folder, self.history) pr_vols = [] for i in range(0, len(valid_dataset), int(len(valid_dataset) / 36)): pr_vols.append(self.model.predict(valid_dataloader[i])) pr_vols = np.concatenate(pr_vols, axis=0) pr_map = map2rgb(np.argmax(pr_vols, axis=-1)) save_images(model_folder + '/pr-{}.png'.format(epoch), pr_map) if epoch == 0: gt_vols, ph_vols = [], [] for i in range(0, len(valid_dataset), int(len(valid_dataset) / 36)): ph_vols.append(valid_dataloader[i][0]) gt_vols.append(valid_dataloader[i][1]) gt_vols = np.concatenate(gt_vols, axis=0) gt_map = map2rgb(np.argmax(gt_vols, axis=-1)) ph_vols = np.concatenate(ph_vols, axis=0) ph_vols = ph_vols.squeeze() save_images(model_folder + '/gt.png'.format(epoch), gt_map) save_images(model_folder + '/img.png'.format(epoch), ph_vols)
def on_epoch_end(self, epoch, logs=None): if logs: for key in logs.keys(): if epoch == 0: self.history[key] = [] self.history[key].append(logs[key]) if epoch % 10 == 0: plot_history_for_callback(model_folder + '/train_history.png', self.history) save_history_for_callback(model_folder, self.history) ph_vols, fl_gts, fl_prs = [], [], [] for i in range(len(valid_dataloader)): ph_vol, fl_gt = valid_dataloader[i] fl_pr = self.model.predict(ph_vol) ph_vols.append(ph_vol) fl_gts.append(fl_gt) fl_prs.append(fl_pr) ph_vols = np.stack(ph_vols).squeeze() fl_gts = np.stack(fl_gts).squeeze() fl_prs = np.stack(fl_prs).squeeze() print('\n max voxel value:{:.2f}'.format(fl_prs.max())) print(ph_vols.shape) save_images(model_folder + '/epoch-{}_ph.png'.format(epoch), ph_vols) save_images(model_folder + '/epoch-{}_gt.png'.format(epoch), fl_gts * 255. / scale) save_images(model_folder + '/epoch-{}_pr.png'.format(epoch), fl_prs * 255. / scale)
def on_epoch_end(self, epoch, logs=None): if logs: for key in logs.keys(): if epoch == 0: self.history[key] = [] self.history[key].append(logs[key]) if epoch % 5 == 0: plot_history_for_callback(model_folder + '/train_history.png', self.history) save_history_for_callback(model_folder, self.history) img_vols, gt_vols, pr_vols = [], [], [] for i in range(0, len(valid_dataset), int(len(valid_dataset) / 36)): img = valid_dataloader[i][0] img_vols.append( np.uint8(255 * (img - img.min()) / (img.max() - img.min()))) gt_vols.append(valid_dataloader[i][1]) pr_vols.append(self.model.predict(valid_dataloader[i])) img_vols = np.concatenate(img_vols, axis=0) gt_vols = np.concatenate(gt_vols, axis=0) gt_map = map2rgb(np.argmax(gt_vols, axis=-1)) pr_vols = np.concatenate(pr_vols, axis=0) pr_map = map2rgb(np.argmax(pr_vols, axis=-1)) save_images(model_folder + '/epoch-{}-img.png'.format(epoch), img_vols) save_images(model_folder + '/epoch-{}-gt.png'.format(epoch), gt_map) save_images(model_folder + '/epoch-{}-pr.png'.format(epoch), pr_map)
def on_epoch_end(self, epoch, logs=None): if logs: for key in logs.keys(): if epoch == 0: self.history[key] = [] self.history[key].append(logs[key]) if epoch % 5 == 0: plot_history_for_callback(model_folder + '/train_history.png', self.history) save_history_for_callback(model_folder, self.history)