def main(flags): nn_utils.set_gpu(GPU) # define network model = unet.UNet(flags.num_classes, flags.patch_size, suffix=flags.model_suffix, learn_rate=flags.learning_rate, decay_step=flags.decay_step, decay_rate=flags.decay_rate, epochs=flags.epochs, batch_size=flags.batch_size) overlap = model.get_overlap() cm = collectionMaker.read_collection(raw_data_path=flags.data_dir, field_name='austin,chicago,kitsap,tyrol-w,vienna', field_id=','.join(str(i) for i in range(37)), rgb_ext='RGB', gt_ext='GT', file_ext='tif', force_run=False, clc_name=flags.ds_name) gt_d255 = collectionEditor.SingleChanMult(cm.clc_dir, 1 / 255, ['GT', 'gt_d255']). \ run(force_run=False, file_ext='png', d_type=np.uint8, ) cm.replace_channel(gt_d255.files, True, ['GT', 'gt_d255']) cm.print_meta_data() file_list_train = cm.load_files(field_id=','.join(str(i) for i in range(6, 37)), field_ext='RGB,gt_d255') file_list_valid = cm.load_files(field_id=','.join(str(i) for i in range(6)), field_ext='RGB,gt_d255') chan_mean = cm.meta_data['chan_mean'][:3] patch_list_train = patchExtractor.PatchExtractor(flags.patch_size, flags.tile_size, flags.ds_name + '_train', overlap, overlap // 2). \ run(file_list=file_list_train, file_exts=['jpg', 'png'], force_run=False).get_filelist() patch_list_valid = patchExtractor.PatchExtractor(flags.patch_size, flags.tile_size, flags.ds_name + '_valid', overlap, overlap // 2). \ run(file_list=file_list_valid, file_exts=['jpg', 'png'], force_run=False).get_filelist() train_init_op, valid_init_op, reader_op = \ dataReaderSegmentation.DataReaderSegmentationTrainValid( flags.patch_size, patch_list_train, patch_list_valid, batch_size=flags.batch_size, chan_mean=chan_mean, aug_func=[reader_utils.image_flipping, reader_utils.image_rotating], random=True, has_gt=True, gt_dim=1, include_gt=True, valid_mult=flags.val_mult).read_op() feature, label = reader_op model.create_graph(feature) model.compile(feature, label, flags.n_train, flags.n_valid, flags.patch_size, ersaPath.PATH['model'], par_dir=flags.model_par_dir, val_mult=flags.val_mult, loss_type='xent') train_hook = hook.ValueSummaryHook(flags.verb_step, [model.loss, model.lr_op], value_names=['train_loss', 'learning_rate'], print_val=[0]) model_save_hook = hook.ModelSaveHook(model.get_epoch_step()*flags.save_epoch, model.ckdir) valid_loss_hook = hook.ValueSummaryHook(model.get_epoch_step(), [model.loss, model.loss_iou], value_names=['valid_loss', 'valid_mIoU'], log_time=True, run_time=model.n_valid, iou_pos=1) image_hook = hook.ImageValidSummaryHook(model.input_size, model.get_epoch_step(), feature, label, model.output, nn_utils.image_summary, img_mean=cm.meta_data['chan_mean']) start_time = time.time() model.train(train_hooks=[train_hook, model_save_hook], valid_hooks=[valid_loss_hook, image_hook], train_init=train_init_op, valid_init=valid_init_op) print('Duration: {:.3f}'.format((time.time() - start_time)/3600))
def main(flags): nn_utils.set_gpu(GPU) # define network model = pspnet.PSPNet(flags.num_classes, flags.tile_size, suffix=flags.model_suffix, learn_rate=flags.learning_rate, decay_step=flags.decay_step, decay_rate=flags.decay_rate, epochs=flags.epochs, batch_size=flags.batch_size, weight_decay=flags.weight_decay, momentum=flags.momentum) cm_train = cityscapes_reader.CollectionMakerCityscapes(flags.data_dir, flags.rgb_type, flags.gt_type, 'train', flags.rgb_ext, flags.gt_ext, ['png', 'png'], clc_name='{}_train'.format(flags.ds_name), force_run=flags.force_run) cm_valid = cityscapes_reader.CollectionMakerCityscapes(flags.data_dir, flags.rgb_type, flags.gt_type, 'val', flags.rgb_ext, flags.gt_ext, ['png', 'png'], clc_name='{}_valid'.format(flags.ds_name), force_run=flags.force_run) cm_train.print_meta_data() cm_valid.print_meta_data() resize_func = lambda img: resize_image(img, flags.tile_size) train_init_op, valid_init_op, reader_op = dataReaderSegmentation.DataReaderSegmentationTrainValid( flags.tile_size, cm_train.meta_data['file_list'], cm_valid.meta_data['file_list'], flags.batch_size, cm_train.meta_data['chan_mean'], aug_func=[reader_utils.image_flipping_hori, reader_utils.image_scaling_with_label], random=True, has_gt=True, gt_dim=1, include_gt=True, valid_mult=flags.val_mult, global_func=resize_func)\ .read_op() feature, label = reader_op model.create_graph(feature) model.load_resnet(flags.res_dir) model.compile(feature, label, flags.n_train, flags.n_valid, flags.tile_size, ersaPath.PATH['model'], par_dir=flags.model_par_dir, val_mult=flags.val_mult, loss_type='xent') train_hook = hook.ValueSummaryHook(flags.verb_step, [model.loss, model.lr_op], value_names=['train_loss', 'learning_rate'], print_val=[0]) model_save_hook = hook.ModelSaveHook(model.get_epoch_step()*flags.save_epoch, model.ckdir) valid_loss_hook = hook.ValueSummaryHook(model.get_epoch_step(), [model.loss, model.loss_iou], value_names=['valid_loss', 'valid_mIoU'], log_time=True, run_time=model.n_valid, iou_pos=1) image_hook = hook.ImageValidSummaryHook(model.input_size, model.get_epoch_step(), feature, label, model.output, cityscapes_labels.image_summary, img_mean=cm_train.meta_data['chan_mean']) start_time = time.time() model.train(train_hooks=[train_hook, model_save_hook], valid_hooks=[valid_loss_hook, image_hook], train_init=train_init_op, valid_init=valid_init_op) print('Duration: {:.3f}'.format((time.time() - start_time)/3600))
hist_match = ga.run(force_run=False, file_list=file_list) cm.add_channel(hist_match.get_files(), '.*rgb_hist') cm.print_meta_data() file_list_train = cm.load_files(field_name='aus10,aus30', field_id='', field_ext='.*rgb_hist,.*gt_d255') file_list_valid = cm.load_files(field_name='aus50', field_id='', field_ext='.*rgb_hist,.*gt_d255') chan_mean = cm.meta_data['chan_mean'][-3:] patch_list_train = patchExtractor.PatchExtractor(patch_size, tile_size, ds_name+'_train', overlap, overlap//2).\ run(file_list=file_list_train, file_exts=['jpg', 'png'], force_run=False).get_filelist() patch_list_valid = patchExtractor.PatchExtractor(patch_size, tile_size, ds_name+'_valid', overlap, overlap//2).\ run(file_list=file_list_valid, file_exts=['jpg', 'png'], force_run=False).get_filelist() train_init_op, valid_init_op, reader_op = \ dataReaderSegmentation.DataReaderSegmentationTrainValid( patch_size, patch_list_train, patch_list_valid, batch_size=bs, chan_mean=chan_mean, aug_func=[reader_utils.image_flipping, reader_utils.image_rotating], random=True, has_gt=True, gt_dim=1, include_gt=True, valid_mult=valid_mult).read_op() feature, label = reader_op unet.create_graph(feature) unet.compile(feature, label, n_train, n_valid, patch_size, ersaPath.PATH['model'], par_dir=ds_name, loss_type='xent') train_hook = hook.ValueSummaryHook(verb_step, [unet.loss, unet.lr_op], value_names=['train_loss', 'learning_rate'], print_val=[0]) model_save_hook = hook.ModelSaveHook(unet.get_epoch_step()*save_epoch, unet.ckdir) valid_loss_hook = hook.ValueSummaryHook(unet.get_epoch_step(), [unet.loss, unet.loss_iou], value_names=['valid_loss', 'IoU'], log_time=True, run_time=unet.n_valid, iou_pos=1) image_hook = hook.ImageValidSummaryHook(unet.input_size, unet.get_epoch_step(), feature, label, unet.pred, nn_utils.image_summary, img_mean=chan_mean) start_time = time.time() unet.load(model_dir)
def main(flags): nn_utils.set_gpu(GPU) # define network model = pspnet.PSPNet(flags.num_classes, flags.patch_size, suffix=flags.model_suffix, learn_rate=flags.learning_rate, decay_step=flags.decay_step, decay_rate=flags.decay_rate, epochs=flags.epochs, batch_size=flags.batch_size, weight_decay=flags.weight_decay, momentum=flags.momentum) overlap = model.get_overlap() cm = collectionMaker.read_collection(raw_data_path=flags.data_dir, field_name='Fresno,Modesto,Stockton', field_id=','.join( str(i) for i in range(663)), rgb_ext='RGB', gt_ext='GT', file_ext='jpg,png', force_run=False, clc_name=flags.ds_name) cm.print_meta_data() file_list_train = cm.load_files(field_id=','.join( str(i) for i in range(0, 250)), field_ext='RGB,GT') file_list_valid = cm.load_files(field_id=','.join( str(i) for i in range(250, 500)), field_ext='RGB,GT') chan_mean = cm.meta_data['chan_mean'][:3] patch_list_train = patchExtractor.PatchExtractor(flags.patch_size, flags.tile_size, flags.ds_name + '_train', overlap, overlap // 2). \ run(file_list=file_list_train, file_exts=['jpg', 'png'], force_run=False).get_filelist() patch_list_valid = patchExtractor.PatchExtractor(flags.patch_size, flags.tile_size, flags.ds_name + '_valid', overlap, overlap // 2). \ run(file_list=file_list_valid, file_exts=['jpg', 'png'], force_run=False).get_filelist() train_init_op, valid_init_op, reader_op = \ dataReaderSegmentation.DataReaderSegmentationTrainValid( flags.patch_size, patch_list_train, patch_list_valid, batch_size=flags.batch_size, chan_mean=chan_mean, aug_func=[reader_utils.image_flipping, reader_utils.image_rotating], random=True, has_gt=True, gt_dim=1, include_gt=True, valid_mult=flags.val_mult).read_op() feature, label = reader_op model.create_graph(feature) model.load_resnet(flags.res_dir, keep_last=False) model.compile(feature, label, flags.n_train, flags.n_valid, flags.patch_size, ersaPath.PATH['model'], par_dir=flags.model_par_dir, val_mult=flags.val_mult, loss_type='xent') train_hook = hook.ValueSummaryHook( flags.verb_step, [model.loss, model.lr_op], value_names=['train_loss', 'learning_rate'], print_val=[0]) model_save_hook = hook.ModelSaveHook( model.get_epoch_step() * flags.save_epoch, model.ckdir) valid_loss_hook = hook.ValueSummaryHookIters( model.get_epoch_step(), [model.loss_xent, model.loss_iou], value_names=['valid_loss', 'valid_mIoU'], log_time=True, run_time=model.n_valid) image_hook = hook.ImageValidSummaryHook(model.input_size, model.get_epoch_step(), feature, label, model.output, nn_utils.image_summary, img_mean=cm.meta_data['chan_mean']) start_time = time.time() model.train(train_hooks=[train_hook, model_save_hook], valid_hooks=[valid_loss_hook, image_hook], train_init=train_init_op, valid_init=valid_init_op) print('Duration: {:.3f}'.format((time.time() - start_time) / 3600))
def main(flags): nn_utils.set_gpu(flags.GPU) np.random.seed(flags.run_id) tf.set_random_seed(flags.run_id) # define network model = unet.UNet(flags.num_classes, flags.patch_size, suffix=flags.suffix, learn_rate=flags.learn_rate, decay_step=flags.decay_step, decay_rate=flags.decay_rate, epochs=flags.epochs, batch_size=flags.batch_size) overlap = model.get_overlap() cm = collectionMaker.read_collection( raw_data_path=flags.data_dir, field_name='Tucson,Colwich,Clyde,Wilmington', field_id=','.join(str(i) for i in range(1, 16)), rgb_ext='RGB', gt_ext='GT', file_ext='tif,png', force_run=False, clc_name=flags.ds_name) gt_d255 = collectionEditor.SingleChanSwitch(cm.clc_dir, { 2: 0, 3: 1, 4: 0, 5: 0, 6: 0, 7: 0 }, ['GT', 'GT_switch'], 'tower_only').run( force_run=False, file_ext='png', d_type=np.uint8, ) cm.replace_channel(gt_d255.files, True, ['GT', 'GT_switch']) cm.print_meta_data() file_list_train = cm.load_files( field_name='Tucson,Colwich,Clyde,Wilmington', field_id=','.join(str(i) for i in range(4, 16)), field_ext='RGB,GT_switch') file_list_valid = cm.load_files( field_name='Tucson,Colwich,Clyde,Wilmington', field_id='1,2,3', field_ext='RGB,GT_switch') patch_list_train = patchExtractor.PatchExtractor(flags.patch_size, ds_name=flags.ds_name + '_tower_only', overlap=overlap, pad=overlap // 2). \ run(file_list=file_list_train, file_exts=['jpg', 'png'], force_run=False).get_filelist() patch_list_valid = patchExtractor.PatchExtractor(flags.patch_size, ds_name=flags.ds_name + '_tower_only', overlap=overlap, pad=overlap // 2). \ run(file_list=file_list_valid, file_exts=['jpg', 'png'], force_run=False).get_filelist() chan_mean = cm.meta_data['chan_mean'] train_init_op, valid_init_op, reader_op = \ dataReaderSegmentation.DataReaderSegmentationTrainValid( flags.patch_size, patch_list_train, patch_list_valid, batch_size=flags.batch_size, chan_mean=chan_mean, aug_func=[reader_utils.image_flipping, reader_utils.image_rotating], random=True, has_gt=True, gt_dim=1, include_gt=True, valid_mult=flags.val_mult).read_op() feature, label = reader_op model.create_graph(feature) model.compile(feature, label, flags.n_train, flags.n_valid, flags.patch_size, ersaPath.PATH['model'], par_dir=flags.par_dir, loss_type='xent', pos_weight=flags.pos_weight) train_hook = hook.ValueSummaryHook( flags.verb_step, [model.loss, model.lr_op], value_names=['train_loss', 'learning_rate'], print_val=[0]) model_save_hook = hook.ModelSaveHook( model.get_epoch_step() * flags.save_epoch, model.ckdir) valid_loss_hook = hook.ValueSummaryHookIters( model.get_epoch_step(), [model.loss_iou, model.loss_xent], value_names=['IoU', 'valid_loss'], log_time=True, run_time=model.n_valid) image_hook = hook.ImageValidSummaryHook(model.input_size, model.get_epoch_step(), feature, label, model.pred, partial( nn_utils.image_summary, label_num=flags.num_classes), img_mean=chan_mean) start_time = time.time() model.train(train_hooks=[train_hook, model_save_hook], valid_hooks=[valid_loss_hook, image_hook], train_init=train_init_op, valid_init=valid_init_op) print('Duration: {:.3f}'.format((time.time() - start_time) / 3600))
def main(flags): nn_utils.set_gpu(flags.GPU) for start_layer in flags.start_layer: if start_layer >= 10: suffix_base = 'aemo_newloss' else: suffix_base = 'aemo_newloss_up{}'.format(start_layer) if flags.from_scratch: suffix_base += '_scratch' for lr in flags.learn_rate: for run_id in range(4): suffix = '{}_{}'.format(suffix_base, run_id) tf.reset_default_graph() np.random.seed(run_id) tf.set_random_seed(run_id) # define network model = unet.UNet(flags.num_classes, flags.patch_size, suffix=suffix, learn_rate=lr, decay_step=flags.decay_step, decay_rate=flags.decay_rate, epochs=flags.epochs, batch_size=flags.batch_size) overlap = model.get_overlap() cm = collectionMaker.read_collection(raw_data_path=flags.data_dir, field_name='aus10,aus30,aus50', field_id='', rgb_ext='.*rgb', gt_ext='.*gt', file_ext='tif', force_run=False, clc_name=flags.ds_name) cm.print_meta_data() file_list_train = cm.load_files(field_name='aus10,aus30', field_id='', field_ext='.*rgb,.*gt') file_list_valid = cm.load_files(field_name='aus50', field_id='', field_ext='.*rgb,.*gt') patch_list_train = patchExtractor.PatchExtractor(flags.patch_size, flags.tile_size, flags.ds_name + '_train_hist', overlap, overlap // 2). \ run(file_list=file_list_train, file_exts=['jpg', 'png'], force_run=False).get_filelist() patch_list_valid = patchExtractor.PatchExtractor(flags.patch_size, flags.tile_size, flags.ds_name + '_valid_hist', overlap, overlap // 2). \ run(file_list=file_list_valid, file_exts=['jpg', 'png'], force_run=False).get_filelist() chan_mean = cm.meta_data['chan_mean'] train_init_op, valid_init_op, reader_op = \ dataReaderSegmentation.DataReaderSegmentationTrainValid( flags.patch_size, patch_list_train, patch_list_valid, batch_size=flags.batch_size, chan_mean=chan_mean, aug_func=[reader_utils.image_flipping, reader_utils.image_rotating], random=True, has_gt=True, gt_dim=1, include_gt=True, valid_mult=flags.val_mult).read_op() feature, label = reader_op model.create_graph(feature) if start_layer >= 10: model.compile(feature, label, flags.n_train, flags.n_valid, flags.patch_size, ersaPath.PATH['model'], par_dir=flags.par_dir, loss_type='xent') else: model.compile(feature, label, flags.n_train, flags.n_valid, flags.patch_size, ersaPath.PATH['model'], par_dir=flags.par_dir, loss_type='xent', train_var_filter=['layerup{}'.format(i) for i in range(start_layer, 10)]) train_hook = hook.ValueSummaryHook(flags.verb_step, [model.loss, model.lr_op], value_names=['train_loss', 'learning_rate'], print_val=[0]) model_save_hook = hook.ModelSaveHook(model.get_epoch_step() * flags.save_epoch, model.ckdir) valid_loss_hook = hook.ValueSummaryHookIters(model.get_epoch_step(), [model.loss_xent, model.loss_iou], value_names=['valid_loss', 'IoU'], log_time=True, run_time=model.n_valid) image_hook = hook.ImageValidSummaryHook(model.input_size, model.get_epoch_step(), feature, label, model.pred, nn_utils.image_summary, img_mean=chan_mean) start_time = time.time() if not flags.from_scratch: model.load(flags.model_dir) model.train(train_hooks=[train_hook, model_save_hook], valid_hooks=[valid_loss_hook, image_hook], train_init=train_init_op, valid_init=valid_init_op) print('Duration: {:.3f}'.format((time.time() - start_time) / 3600))
def main(flags): nn_utils.set_gpu(flags.GPU) for start_layer in flags.start_layer: if start_layer >= 10: suffix_base = 'aemo' else: suffix_base = 'aemo_up{}'.format(start_layer) if flags.from_scratch: suffix_base += '_scratch' for lr in flags.learn_rate: for run_id in range(1): suffix = '{}_{}'.format(suffix_base, run_id) tf.reset_default_graph() np.random.seed(run_id) tf.set_random_seed(run_id) # define network model = unet.UNet(flags.num_classes, flags.patch_size, suffix=suffix, learn_rate=lr, decay_step=flags.decay_step, decay_rate=flags.decay_rate, epochs=flags.epochs, batch_size=flags.batch_size) file_list = os.path.join(flags.data_dir, 'file_list.txt') lines = ersa_utils.load_file(file_list) patch_list_train = [] patch_list_valid = [] train_tile_names = ['aus10', 'aus30'] valid_tile_names = ['aus50'] for line in lines: tile_name = os.path.basename( line.split(' ')[0]).split('_')[0].strip() if tile_name in train_tile_names: patch_list_train.append(line.strip().split(' ')) elif tile_name in valid_tile_names: patch_list_valid.append(line.strip().split(' ')) else: raise ValueError cm = collectionMaker.read_collection('aemo_align') chan_mean = cm.meta_data['chan_mean'] train_init_op, valid_init_op, reader_op = \ dataReaderSegmentation.DataReaderSegmentationTrainValid( flags.patch_size, patch_list_train, patch_list_valid, batch_size=flags.batch_size, chan_mean=chan_mean, aug_func=[reader_utils.image_flipping, reader_utils.image_rotating], random=True, has_gt=True, gt_dim=1, include_gt=True, valid_mult=flags.val_mult).read_op() feature, label = reader_op model.create_graph(feature) if start_layer >= 10: model.compile(feature, label, flags.n_train, flags.n_valid, flags.patch_size, ersaPath.PATH['model'], par_dir=flags.par_dir, loss_type='xent') else: model.compile(feature, label, flags.n_train, flags.n_valid, flags.patch_size, ersaPath.PATH['model'], par_dir=flags.par_dir, loss_type='xent', train_var_filter=[ 'layerup{}'.format(i) for i in range(start_layer, 10) ]) train_hook = hook.ValueSummaryHook( flags.verb_step, [model.loss, model.lr_op], value_names=['train_loss', 'learning_rate'], print_val=[0]) model_save_hook = hook.ModelSaveHook( model.get_epoch_step() * flags.save_epoch, model.ckdir) valid_loss_hook = hook.ValueSummaryHookIters( model.get_epoch_step(), [model.loss_xent, model.loss_iou], value_names=['valid_loss', 'IoU'], log_time=True, run_time=model.n_valid) image_hook = hook.ImageValidSummaryHook(model.input_size, model.get_epoch_step(), feature, label, model.pred, nn_utils.image_summary, img_mean=chan_mean) start_time = time.time() if not flags.from_scratch: model.load(flags.model_dir) model.train(train_hooks=[train_hook, model_save_hook], valid_hooks=[valid_loss_hook, image_hook], train_init=train_init_op, valid_init=valid_init_op) print('Duration: {:.3f}'.format( (time.time() - start_time) / 3600))