def main(_argv): class_names = [c.strip() for c in open("datasets/classes.txt").readlines()] dataset = load_tfrecord_dataset(input_filename, 416, 416, image_type='jpg') idx = 0 for image, labels in dataset: idx += 1 # boxes = [] # scores = [] # classes = [] # for x1, y1, x2, y2, label in labels: # if x1 == 0 and x2 == 0: # continue # boxes.append((x1, y1, x2, y2)) # scores.append(1) # classes.append(label) # nums = [len(boxes)] # boxes = [boxes] # scores = [scores] # classes = [classes] # img = cv2.cvtColor(image.numpy(), cv2.COLOR_RGB2BGR) if idx % 100 == 0: print(idx) # img = cv2.resize(img, (1088, 612), interpolation=cv2.INTER_LANCZOS4) # img = draw_outputs(img, (boxes, scores, classes, nums), class_names) # cv2.imwrite(f"buggy/{idx}.jpg", img) print(idx)
def main(_argv): # 读取 类别 并生成 list class_names = [c.strip() for c in open(FLAGS.classes).readlines()] logging.info('classes loaded') # 读取 TF record格式 的数据集并 随机排序 dataset = load_tfrecord_dataset(FLAGS.dataset, FLAGS.classes, FLAGS.size) dataset = dataset.shuffle(512) # 读取第一张图片并可视化并保存 for image, labels in dataset.take(1): boxes = [] scores = [] classes = [] for x1, y1, x2, y2, label in labels: if x1 == 0 and x2 == 0: continue boxes.append((x1, y1, x2, y2)) scores.append(1) classes.append(label) nums = [len(boxes)] boxes = [boxes] scores = [scores] classes = [classes] logging.info('labels:') for i in range(nums[0]): logging.info('\t{}, {}, {}'.format(class_names[int(classes[0][i])], np.array(scores[0][i]), np.array(boxes[0][i]))) img = cv2.cvtColor(image.numpy(), cv2.COLOR_RGB2BGR) img = draw_outputs(img, (boxes, scores, classes, nums), class_names) cv2.imwrite(FLAGS.output, img) logging.info('output saved to: {}'.format(FLAGS.output))
def main(_argv): # 打开内存增长 physical_devices = tf.config.experimental.list_physical_devices('GPU') for physical_device in physical_devices: tf.config.experimental.set_memory_growth(physical_device, True) # 创建网络模型 if FLAGS.tiny: yolo = YoloV3Tiny(classes=FLAGS.num_classes) else: yolo = YoloV3(classes=FLAGS.num_classes) # 加载权重 yolo.load_weights(FLAGS.weights).expect_partial() logging.info('weights loaded') # 加载类别列表 class_names = [c.strip() for c in open(FLAGS.classes).readlines()] logging.info('classes loaded') # 读取图片数据 if FLAGS.tfrecord: dataset = load_tfrecord_dataset(FLAGS.tfrecord, FLAGS.classes, FLAGS.size) dataset = dataset.shuffle(512) img_raw, _label = next(iter(dataset.take(1))) else: img_raw = tf.image.decode_image(open(FLAGS.image, 'rb').read(), channels=3) # 调整图片 img = tf.expand_dims(img_raw, 0) img = transform_images(img, FLAGS.size) # 对图片进行目标检测 t1 = time.time() boxes, scores, classes, nums = yolo(img) t2 = time.time() logging.info('time: {}'.format(t2 - t1)) # 可视化输出 logging.info('detections:') for i in range(nums[0]): logging.info('\t{}, {}, {}'.format(class_names[int(classes[0][i])], np.array(scores[0][i]), np.array(boxes[0][i]))) img = cv2.cvtColor(img_raw.numpy(), cv2.COLOR_RGB2BGR) img = draw_outputs(img, (boxes, scores, classes, nums), class_names) cv2.imwrite(FLAGS.output, img) logging.info('output saved to: {}'.format(FLAGS.output))
def main(_argv): physical_devices = tf.config.experimental.list_physical_devices('GPU') for physical_device in physical_devices: tf.config.experimental.set_memory_growth(physical_device, True) if FLAGS.tiny: model = YoloV3Tiny(FLAGS.size, training=True, classes=FLAGS.num_classes) anchors = yolo_tiny_anchors anchor_masks = yolo_tiny_anchor_masks else: model = YoloV3(FLAGS.size, training=True, classes=FLAGS.num_classes) anchors = yolo_anchors anchor_masks = yolo_anchor_masks train_dataset = dataset.load_fake_dataset() if FLAGS.dataset: train_dataset = dataset.load_tfrecord_dataset(FLAGS.dataset, FLAGS.classes, FLAGS.size) train_dataset = train_dataset.shuffle(buffer_size=512) train_dataset = train_dataset.batch(FLAGS.batch_size) train_dataset = train_dataset.map(lambda x, y: ( dataset.transform_images(x, FLAGS.size), dataset.transform_targets(y, anchors, anchor_masks, FLAGS.size))) train_dataset = train_dataset.prefetch( buffer_size=tf.data.experimental.AUTOTUNE) val_dataset = dataset.load_fake_dataset() if FLAGS.val_dataset: val_dataset = dataset.load_tfrecord_dataset(FLAGS.val_dataset, FLAGS.classes, FLAGS.size) val_dataset = val_dataset.batch(FLAGS.batch_size) val_dataset = val_dataset.map(lambda x, y: ( dataset.transform_images(x, FLAGS.size), dataset.transform_targets(y, anchors, anchor_masks, FLAGS.size))) # Configure the model for transfer learning if FLAGS.transfer == 'none': pass # Nothing to do elif FLAGS.transfer in ['darknet', 'no_output']: # Darknet transfer is a special case that works # with incompatible number of classes # reset top layers if FLAGS.tiny: model_pretrained = YoloV3Tiny(FLAGS.size, training=True, classes=FLAGS.weights_num_classes or FLAGS.num_classes) else: model_pretrained = YoloV3(FLAGS.size, training=True, classes=FLAGS.weights_num_classes or FLAGS.num_classes) model_pretrained.load_weights(FLAGS.weights) if FLAGS.transfer == 'darknet': model.get_layer('yolo_darknet').set_weights( model_pretrained.get_layer('yolo_darknet').get_weights()) freeze_all(model.get_layer('yolo_darknet')) elif FLAGS.transfer == 'no_output': for l in model.layers: if not l.name.startswith('yolo_output'): l.set_weights( model_pretrained.get_layer(l.name).get_weights()) freeze_all(l) else: # All other transfer require matching classes model.load_weights(FLAGS.weights) if FLAGS.transfer == 'fine_tune': # freeze darknet and fine tune other layers darknet = model.get_layer('yolo_darknet') freeze_all(darknet) elif FLAGS.transfer == 'frozen': # freeze everything freeze_all(model) optimizer = tf.keras.optimizers.Adam(lr=FLAGS.learning_rate) loss = [ YoloLoss(anchors[mask], classes=FLAGS.num_classes) for mask in anchor_masks ] if FLAGS.mode == 'eager_tf': # Eager mode is great for debugging # Non eager graph mode is recommended for real training avg_loss = tf.keras.metrics.Mean('loss', dtype=tf.float32) avg_val_loss = tf.keras.metrics.Mean('val_loss', dtype=tf.float32) for epoch in range(1, FLAGS.epochs + 1): for batch, (images, labels) in enumerate(train_dataset): with tf.GradientTape() as tape: outputs = model(images, training=True) regularization_loss = tf.reduce_sum(model.losses) pred_loss = [] for output, label, loss_fn in zip(outputs, labels, loss): pred_loss.append(loss_fn(label, output)) total_loss = tf.reduce_sum(pred_loss) + regularization_loss grads = tape.gradient(total_loss, model.trainable_variables) optimizer.apply_gradients(zip(grads, model.trainable_variables)) logging.info("{}_train_{}, {}, {}".format( epoch, batch, total_loss.numpy(), list(map(lambda x: np.sum(x.numpy()), pred_loss)))) avg_loss.update_state(total_loss) for batch, (images, labels) in enumerate(val_dataset): outputs = model(images) regularization_loss = tf.reduce_sum(model.losses) pred_loss = [] for output, label, loss_fn in zip(outputs, labels, loss): pred_loss.append(loss_fn(label, output)) total_loss = tf.reduce_sum(pred_loss) + regularization_loss logging.info("{}_val_{}, {}, {}".format( epoch, batch, total_loss.numpy(), list(map(lambda x: np.sum(x.numpy()), pred_loss)))) avg_val_loss.update_state(total_loss) logging.info("{}, train: {}, val: {}".format( epoch, avg_loss.result().numpy(), avg_val_loss.result().numpy())) avg_loss.reset_states() avg_val_loss.reset_states() model.save_weights('checkpoints/yolov3_train_{}.tf'.format(epoch)) else: model.compile(optimizer=optimizer, loss=loss, run_eagerly=(FLAGS.mode == 'eager_fit')) callbacks = [ ReduceLROnPlateau(verbose=1), EarlyStopping(patience=3, verbose=1), ModelCheckpoint('checkpoints/yolov3_train_{epoch}.tf', verbose=1, save_weights_only=True), TensorBoard(log_dir='logs') ] history = model.fit(train_dataset, epochs=FLAGS.epochs, callbacks=callbacks, validation_data=val_dataset)
def main(_argv): # 打开内存增长 physical_devices = tf.config.experimental.list_physical_devices('GPU') for physical_device in physical_devices: tf.config.experimental.set_memory_growth(physical_device, True) # 创建网络模型 if FLAGS.tiny: model = YoloV3Tiny(FLAGS.size, training=True, classes=FLAGS.num_classes) anchors = yolo_tiny_anchors anchor_masks = yolo_tiny_anchor_masks else: model = YoloV3(FLAGS.size, training=True, classes=FLAGS.num_classes) anchors = yolo_anchors anchor_masks = yolo_anchor_masks # 加载训练集 if FLAGS.dataset: train_dataset = dataset.load_tfrecord_dataset(FLAGS.dataset, FLAGS.classes, FLAGS.size) else: train_dataset = dataset.load_fake_dataset() # 对训练集进行 随机化 取数据 做映射 预加载 train_dataset = train_dataset.shuffle(buffer_size=512) train_dataset = train_dataset.batch(FLAGS.batch_size) train_dataset = train_dataset.map(lambda x, y: ( dataset.transform_images(x, FLAGS.size), dataset.transform_targets(y, anchors, anchor_masks, FLAGS.size))) train_dataset = train_dataset.prefetch( buffer_size=tf.data.experimental.AUTOTUNE) # 加载验证集 if FLAGS.val_dataset: val_dataset = dataset.load_tfrecord_dataset(FLAGS.val_dataset, FLAGS.classes, FLAGS.size) else: val_dataset = dataset.load_fake_dataset() # 对验证集进行 取数据 做映射 val_dataset = val_dataset.batch(FLAGS.batch_size) val_dataset = val_dataset.map(lambda x, y: ( dataset.transform_images(x, FLAGS.size), dataset.transform_targets(y, anchors, anchor_masks, FLAGS.size))) # 配置迁移学习 if FLAGS.transfer == 'none': pass elif FLAGS.transfer in ['darknet', 'no_output']: # darknet 迁移学习可以在类别数量不兼容的情况下工作 # 重置顶层 if FLAGS.tiny: model_pretrained = YoloV3Tiny(FLAGS.size, training=True, classes=FLAGS.weights_num_classes or FLAGS.num_classes) else: model_pretrained = YoloV3(FLAGS.size, training=True, classes=FLAGS.weights_num_classes or FLAGS.num_classes) model_pretrained.load_weights(FLAGS.weights) # 固化 darknet if FLAGS.transfer == 'darknet': model.get_layer('yolo_darknet').set_weights( model_pretrained.get_layer('yolo_darknet').get_weights()) freeze_all(model.get_layer('yolo_darknet')) # 固化输出层以外的层 elif FLAGS.transfer == 'no_output': for layer in model.layers: if not layer.name.startswith('yolo_output'): layer.set_weights( model_pretrained.get_layer(layer.name).get_weights()) freeze_all(layer) else: # 其他类型的迁移学习需要正确的类别数 model.load_weights(FLAGS.weights) if FLAGS.transfer == 'fine_tune': # 固化 darknet 并微调其他层 darknet = model.get_layer('yolo_darknet') freeze_all(darknet) elif FLAGS.transfer == 'frozen': # 固化全部 freeze_all(model) optimizer = tf.keras.optimizers.Adam(lr=FLAGS.learning_rate) loss = [ YoloLoss(anchors[mask], classes=FLAGS.num_classes) for mask in anchor_masks ] # 是否使用 Eager Execution if FLAGS.mode == 'eager_tf': # Eager Execution 模式可以在运行过程中查看,方便调试 avg_loss = tf.keras.metrics.Mean('loss', dtype=tf.float32) avg_val_loss = tf.keras.metrics.Mean('val_loss', dtype=tf.float32) for epoch in range(1, FLAGS.epochs + 1): # 使用梯度下降法用自动微分优化训练集的损失 for batch, (images, labels) in enumerate(train_dataset): with tf.GradientTape() as tape: outputs = model(images, training=True) regularization_loss = tf.reduce_sum(model.losses) pred_loss = [] for output, label, loss_fn in zip(outputs, labels, loss): pred_loss.append(loss_fn(label, output)) total_loss = tf.reduce_sum(pred_loss) + regularization_loss grads = tape.gradient(total_loss, model.trainable_variables) optimizer.apply_gradients(zip(grads, model.trainable_variables)) logging.info("{}_train_{}, {}, {}".format( epoch, batch, total_loss.numpy(), list(map(lambda x: np.sum(x.numpy()), pred_loss)))) avg_loss.update_state(total_loss) # 计算验证集的损失 for batch, (images, labels) in enumerate(val_dataset): outputs = model(images) regularization_loss = tf.reduce_sum(model.losses) pred_loss = [] for output, label, loss_fn in zip(outputs, labels, loss): pred_loss.append(loss_fn(label, output)) total_loss = tf.reduce_sum(pred_loss) + regularization_loss logging.info("{}_val_{}, {}, {}".format( epoch, batch, total_loss.numpy(), list(map(lambda x: np.sum(x.numpy()), pred_loss)))) avg_val_loss.update_state(total_loss) logging.info("{}, train: {}, val: {}".format( epoch, avg_loss.result().numpy(), avg_val_loss.result().numpy())) avg_loss.reset_states() avg_val_loss.reset_states() model.save_weights('checkpoints/yolov3_train_{}.tf'.format(epoch)) else: model.compile(optimizer=optimizer, loss=loss, run_eagerly=(FLAGS.mode == 'eager_fit')) callbacks = [ ReduceLROnPlateau(verbose=1), EarlyStopping(patience=3, verbose=1), ModelCheckpoint('checkpoints/yolov3_train_{epoch}.tf', verbose=1, save_weights_only=True), TensorBoard(log_dir='logs') ] history = model.fit(train_dataset, epochs=FLAGS.epochs, callbacks=callbacks, validation_data=val_dataset) print(history.history)