def dataset_creator(config): opt = config["opt"] hyper_params = config["hyper_params"] train_data, _ = data_utils.get_dataset("voc/2007", "train+validation") val_data, _ = data_utils.get_dataset("voc/2007", "test") if opt.with_voc12: voc_2012_data, _ = data_utils.get_dataset("voc/2012", "train+validation") train_data = train_data.concatenate(voc_2012_data) img_size = hyper_params["img_size"] train_data = train_data.map(lambda x: data_utils.preprocessing( x, img_size, img_size, augmentation.apply)) val_data = val_data.map( lambda x: data_utils.preprocessing(x, img_size, img_size)) data_shapes = data_utils.get_data_shapes() padding_values = data_utils.get_padding_values() train_data = train_data.shuffle(opt.batch_size * 4).padded_batch( opt.batch_size, padded_shapes=data_shapes, padding_values=padding_values) val_data = val_data.padded_batch(opt.batch_size, padded_shapes=data_shapes, padding_values=padding_values) prior_boxes = bbox_utils.generate_prior_boxes( hyper_params["feature_map_shapes"], hyper_params["aspect_ratios"]) ssd_train_feed = train_utils.generator(train_data, prior_boxes, hyper_params) ssd_val_feed = train_utils.generator(val_data, prior_boxes, hyper_params) return ssd_train_feed, ssd_val_feed
hyper_params["total_labels"] = len(labels) img_size = hyper_params["img_size"] data_types = data_utils.get_data_types() data_shapes = data_utils.get_data_shapes() padding_values = data_utils.get_padding_values() if use_custom_images: img_paths = data_utils.get_custom_imgs(custom_image_path) total_items = len(img_paths) test_data = tf.data.Dataset.from_generator( lambda: data_utils.custom_data_generator(img_paths, img_size, img_size ), data_types, data_shapes) else: test_data = test_data.map( lambda x: data_utils.preprocessing(x, img_size, img_size)) test_data = test_data.padded_batch(batch_size, padded_shapes=data_shapes, padding_values=padding_values) rpn_model, _ = get_model(hyper_params) frcnn_model_path = io_utils.get_model_path("faster_rcnn", backbone) rpn_model_path = io_utils.get_model_path("rpn", backbone) model_path = frcnn_model_path if load_weights_from_frcnn else rpn_model_path rpn_model.load_weights(model_path, by_name=True) anchors = bbox_utils.generate_anchors(hyper_params) for image_data in test_data:
"voc/2012", "train+validation") voc_2012_total_items = data_utils.get_total_item_size( voc_2012_info, "train+validation") train_total_items += voc_2012_total_items train_data = train_data.concatenate(voc_2012_data) # # Get labels labels = data_utils.get_labels(info) # Add background label into labels labels = ["bg"] + labels # Get hyper-parameters and image size hyper_params["total_labels"] = len(labels) img_size = hyper_params["img_size"] # Data pre-processing train_data = train_data.map(lambda x: data_utils.preprocessing( x, img_size, img_size, augmentation.apply)) val_data = val_data.map( lambda x: data_utils.preprocessing(x, img_size, img_size)) data_shapes = data_utils.get_data_shapes() padding_values = data_utils.get_padding_values() train_data = train_data.shuffle(batch_size * 4).padded_batch( batch_size, padded_shapes=data_shapes, padding_values=padding_values) val_data = val_data.padded_batch(batch_size, padded_shapes=data_shapes, padding_values=padding_values) # Setup training model (ssd+vgg) and loss function (location + confidence) ssd_model = get_model(hyper_params) ssd_custom_losses = CustomLoss(hyper_params["neg_pos_ratio"], hyper_params["loc_loss_alpha"]) ssd_model.compile(
val_total_items = data_utils.get_total_item_size(dataset_info, "test") if with_voc_2012: voc_2012_data, voc_2012_info = data_utils.get_dataset( "voc/2012", "train+validation") voc_2012_total_items = data_utils.get_total_item_size( voc_2012_info, "train+validation") train_total_items += voc_2012_total_items train_data = train_data.concatenate(voc_2012_data) labels = data_utils.get_labels(dataset_info) # We add 1 class for background hyper_params["total_labels"] = len(labels) + 1 (img_width, img_height) = hyper_params["img_size"] train_data = train_data.map(lambda x: data_utils.preprocessing( x, img_width, img_height, apply_augmentation=True)) val_data = val_data.map( lambda x: data_utils.preprocessing(x, img_width, img_height)) data_shapes = data_utils.get_data_shapes() padding_values = data_utils.get_padding_values() train_data = train_data.padded_batch(batch_size, padded_shapes=data_shapes, padding_values=padding_values) val_data = val_data.padded_batch(batch_size, padded_shapes=data_shapes, padding_values=padding_values) anchors = bbox_utils.generate_anchors(hyper_params) rpn_train_feed = train_utils.rpn_generator(train_data, anchors, hyper_params) rpn_val_feed = train_utils.rpn_generator(val_data, anchors, hyper_params)
labels = ["bg"] + labels hyper_params["total_labels"] = len(labels) img_size = hyper_params["img_size"] data_types = data_utils.get_data_types() data_shapes = data_utils.get_data_shapes() padding_values = data_utils.get_padding_values() if use_custom_images: img_paths = data_utils.get_custom_imgs(custom_image_path) total_items = len(img_paths) test_data = tf.data.Dataset.from_generator( lambda: data_utils.custom_data_generator(img_paths, img_size, img_size ), data_types, data_shapes) else: test_data = test_data.map(lambda x: data_utils.preprocessing( x, img_size, img_size, evaluate=evaluate)) test_data = test_data.padded_batch(batch_size, padded_shapes=data_shapes, padding_values=padding_values) ssd_model = get_model(hyper_params) ssd_model_path = io_utils.get_model_path(backbone) ssd_model.load_weights(ssd_model_path) prior_boxes = bbox_utils.generate_prior_boxes( hyper_params["feature_map_shapes"], hyper_params["aspect_ratios"]) ssd_decoder_model = get_decoder_model(ssd_model, prior_boxes, hyper_params) step_size = train_utils.get_step_size(total_items, batch_size) pred_bboxes, pred_labels, pred_scores = ssd_decoder_model.predict(
val_data, _ = data_utils.get_dataset("voc/2007", "test") train_total_items = data_utils.get_total_item_size(dataset_info, "train+validation") val_total_items = data_utils.get_total_item_size(dataset_info, "test") if with_voc_2012: voc_2012_data, voc_2012_info = data_utils.get_dataset("voc/2012", "train+validation") voc_2012_total_items = data_utils.get_total_item_size(voc_2012_info, "train+validation") train_total_items += voc_2012_total_items train_data = train_data.concatenate(voc_2012_data) labels = data_utils.get_labels(dataset_info) # We add 1 class for background hyper_params["total_labels"] = len(labels) + 1 # img_size = hyper_params["img_size"] train_data = train_data.map(lambda x: data_utils.preprocessing(x, img_size, img_size, apply_augmentation=True)) val_data = val_data.map(lambda x: data_utils.preprocessing(x, img_size, img_size)) data_shapes = data_utils.get_data_shapes() padding_values = data_utils.get_padding_values() train_data = train_data.padded_batch(batch_size, padded_shapes=data_shapes, padding_values=padding_values) val_data = val_data.padded_batch(batch_size, padded_shapes=data_shapes, padding_values=padding_values) anchors = bbox_utils.generate_anchors(hyper_params) frcnn_train_feed = train_utils.faster_rcnn_generator(train_data, anchors, hyper_params) frcnn_val_feed = train_utils.faster_rcnn_generator(val_data, anchors, hyper_params) # rpn_model, feature_extractor = get_rpn_model(hyper_params) frcnn_model = faster_rcnn.get_model(feature_extractor, rpn_model, anchors, hyper_params) frcnn_model.compile(optimizer=tf.optimizers.Adam(learning_rate=1e-5), loss=[None] * len(frcnn_model.output))