How to run 2 version of tensorflow at the same time

Viewed 23

I have a project which I need to mix 2 programs into one.

Yolov4 running on tensorflow 2.x which is used to detect object in a video and Person-ReID running on tensorflow 1.x which is used to check if 2 personnes are the same or not.

After converting tensorflow 1.x into 2.x using tensorflow.compat.v1 on the majority of function, it runs perfectly.

Now comes the mix of the two programs.

#Import yolov4 library
import time
import tensorflow as tf
physical_devices = tf.config.experimental.list_physical_devices('GPU')
if len(physical_devices) > 0:
    tf.config.experimental.set_memory_growth(physical_devices[0], True)
from absl import app, flags, logging
from absl.flags import FLAGS
import core.utils as utils
from core.yolov4 import filter_boxes
from tensorflow.python.saved_model import tag_constants
from PIL import Image
import cv2
import numpy as np
from imutils.video import FPS
from tensorflow.compat.v1 import ConfigProto
from tensorflow.compat.v1 import InteractiveSession

#Import reid unique library
import cuhk03_dataset

#Flag Yolov4
flags.DEFINE_string('framework', 'tf', '(tf, tflite, trt')
flags.DEFINE_string('weights', './checkpoints/yolov4-416',
                    'path to weights file')
flags.DEFINE_integer('size', 416, 'resize images to')
flags.DEFINE_boolean('tiny', False, 'yolo or yolo-tiny')
flags.DEFINE_string('model', 'yolov4', 'yolov3 or yolov4')
flags.DEFINE_string('video', './data/road.mp4', 'path to input video')
flags.DEFINE_float('iou', 0.45, 'iou threshold')
flags.DEFINE_float('score', 0.25, 'score threshold')
flags.DEFINE_string('output', None, 'path to output video')
flags.DEFINE_string('output_format', 'XVID', 'codec used in VideoWriter when saving video to file')
flags.DEFINE_boolean('dis_cv2_window', False, 'disable cv2 window during the process') # this is good for the .ipynb

#Flag reid
flags.DEFINE_integer('batch_size', '150', 'batch size for training')
flags.DEFINE_integer('max_steps', '1000', 'max steps for training')
flags.DEFINE_string('logs_dir', 'logs/', 'path to logs directory')
flags.DEFINE_string('data_dir', 'data/', 'path to dataset')

#Reid code
IMAGE_WIDTH = 60
IMAGE_HEIGHT = 160
image = []
#Reid preprocess
def preprocess(images, is_train):
    def train():
        split = tf.split(images, [1, 1])
        shape = [1 for _ in list(range(split[0].get_shape()[1]))]
        for i in list(range(len(split))):
            split[i] = tf.reshape(split[i], [FLAGS.batch_size, IMAGE_HEIGHT, IMAGE_WIDTH, 3])
            split[i] = tf.image.resize(split[i], [IMAGE_HEIGHT + 8, IMAGE_WIDTH + 3])
            split[i] = tf.split(split[i], shape)
            for j in list(range(len(split[i]))):
                split[i][j] = tf.reshape(split[i][j], [IMAGE_HEIGHT + 8, IMAGE_WIDTH + 3, 3])
                split[i][j] = tf.image.random_crop(split[i][j], [IMAGE_HEIGHT, IMAGE_WIDTH, 3])
                split[i][j] = tf.image.random_flip_left_right(split[i][j])
                split[i][j] = tf.image.random_brightness(split[i][j], max_delta=32. / 255.)
                split[i][j] = tf.image.random_saturation(split[i][j], lower=0.5, upper=1.5)
                split[i][j] = tf.image.random_hue(split[i][j], max_delta=0.2)
                split[i][j] = tf.image.random_contrast(split[i][j], lower=0.5, upper=1.5)
                split[i][j] = tf.image.per_image_standardization(split[i][j])
        return [tf.reshape(tf.concat(split[0], axis=0), [FLAGS.batch_size, IMAGE_HEIGHT, IMAGE_WIDTH, 3]),
            tf.reshape(tf.concat(split[1], axis=0), [FLAGS.batch_size, IMAGE_HEIGHT, IMAGE_WIDTH, 3])]
    def val():
        split = tf.split(images, [1, 1])
        shape = [1 for _ in list(range(split[0].get_shape()[1]))]
        for i in list(range(len(split))):
            split[i] = tf.reshape(split[i], [FLAGS.batch_size, IMAGE_HEIGHT, IMAGE_WIDTH, 3])
            split[i] = tf.image.resize(split[i], [IMAGE_HEIGHT, IMAGE_WIDTH])
            split[i] = tf.split(split[i], shape)
            for j in list(range(len(split[i]))):
                split[i][j] = tf.reshape(split[i][j], [IMAGE_HEIGHT, IMAGE_WIDTH, 3])
                split[i][j] = tf.image.per_image_standardization(split[i][j])
        return [tf.reshape(tf.concat(split[0], axis=0), [FLAGS.batch_size, IMAGE_HEIGHT, IMAGE_WIDTH, 3]),
            tf.reshape(tf.concat(split[1], axis=0), [FLAGS.batch_size, IMAGE_HEIGHT, IMAGE_WIDTH, 3])]
    return tf.cond(pred=is_train, true_fn=train, false_fn=val)
#Reid network
def network(images1, images2, weight_decay):
    with tf.compat.v1.variable_scope('network'):
        # Tied Convolution
        conv1_1 = tf.compat.v1.layers.conv2d(images1, 20, [5, 5], activation=tf.nn.relu,
            kernel_regularizer=tf.keras.regularizers.l2(0.5 * (weight_decay)), name='conv1_1')
        pool1_1 = tf.compat.v1.layers.max_pooling2d(conv1_1, [2, 2], [2, 2], name='pool1_1')
        conv1_2 = tf.compat.v1.layers.conv2d(pool1_1, 25, [5, 5], activation=tf.nn.relu,
            kernel_regularizer=tf.keras.regularizers.l2(0.5 * (weight_decay)), name='conv1_2')
        pool1_2 = tf.compat.v1.layers.max_pooling2d(conv1_2, [2, 2], [2, 2], name='pool1_2')
        conv2_1 = tf.compat.v1.layers.conv2d(images2, 20, [5, 5], activation=tf.nn.relu,
            kernel_regularizer=tf.keras.regularizers.l2(0.5 * (weight_decay)), name='conv2_1')
        pool2_1 = tf.compat.v1.layers.max_pooling2d(conv2_1, [2, 2], [2, 2], name='pool2_1')
        conv2_2 = tf.compat.v1.layers.conv2d(pool2_1, 25, [5, 5], activation=tf.nn.relu,
            kernel_regularizer=tf.keras.regularizers.l2(0.5 * (weight_decay)), name='conv2_2')
        pool2_2 = tf.compat.v1.layers.max_pooling2d(conv2_2, [2, 2], [2, 2], name='pool2_2')
        # Cross-Input Neighborhood Differences
        trans = tf.transpose(a=pool1_2, perm=[0, 3, 1, 2])
        shape = trans.get_shape().as_list()
        m1s = tf.ones([shape[0], shape[1], shape[2], shape[3], 5, 5])
        reshape = tf.reshape(trans, [shape[0], shape[1], shape[2], shape[3], 1, 1])
        f = tf.multiply(reshape, m1s)

        trans = tf.transpose(a=pool2_2, perm=[0, 3, 1, 2])
        reshape = tf.reshape(trans, [1, shape[0], shape[1], shape[2], shape[3]])
        g = []
        pad = tf.pad(tensor=reshape, paddings=[[0, 0], [0, 0], [0, 0], [2, 2], [2, 2]])
        for i in list(range(shape[2])):
            for j in list(range(shape[3])):
                g.append(pad[:,:,:,i:i+5,j:j+5])

        concat = tf.concat(g, axis=0)
        reshape = tf.reshape(concat, [shape[2], shape[3], shape[0], shape[1], 5, 5])
        g = tf.transpose(a=reshape, perm=[2, 3, 0, 1, 4, 5])
        reshape1 = tf.reshape(tf.subtract(f, g), [shape[0], shape[1], shape[2] * 5, shape[3] * 5])
        reshape2 = tf.reshape(tf.subtract(g, f), [shape[0], shape[1], shape[2] * 5, shape[3] * 5])
        k1 = tf.nn.relu(tf.transpose(a=reshape1, perm=[0, 2, 3, 1]), name='k1')
        k2 = tf.nn.relu(tf.transpose(a=reshape2, perm=[0, 2, 3, 1]), name='k2')

        # Patch Summary Features
        l1 = tf.compat.v1.layers.conv2d(k1, 25, [5, 5], (5, 5), activation=tf.nn.relu,
            kernel_regularizer=tf.keras.regularizers.l2(0.5 * (weight_decay)), name='l1')
        l2 = tf.compat.v1.layers.conv2d(k2, 25, [5, 5], (5, 5), activation=tf.nn.relu,
            kernel_regularizer=tf.keras.regularizers.l2(0.5 * (weight_decay)), name='l2')

        # Across-Patch Features
        m1 = tf.compat.v1.layers.conv2d(l1, 25, [3, 3], activation=tf.nn.relu,
            kernel_regularizer=tf.keras.regularizers.l2(0.5 * (weight_decay)), name='m1')
        pool_m1 = tf.compat.v1.layers.max_pooling2d(m1, [2, 2], [2, 2], padding='same', name='pool_m1')
        m2 = tf.compat.v1.layers.conv2d(l2, 25, [3, 3], activation=tf.nn.relu,
            kernel_regularizer=tf.keras.regularizers.l2(0.5 * (weight_decay)), name='m2')
        pool_m2 = tf.compat.v1.layers.max_pooling2d(m2, [2, 2], [2, 2], padding='same', name='pool_m2')

        # Higher-Order Relationships
        concat = tf.concat([pool_m1, pool_m2], axis=3)
        reshape = tf.reshape(concat, [FLAGS.batch_size, -1])
        fc1 = tf.compat.v1.layers.dense(reshape, 500, tf.nn.relu, name='fc1')
        fc2 = tf.compat.v1.layers.dense(fc1, 2, name='fc2')

        return fc2

#Yolo main function
def yolo(_argv):
    print('\nExecuting Yolo function\n')

    config = ConfigProto()
    config.gpu_options.allow_growth = True
    session = InteractiveSession(config=config)
    STRIDES, ANCHORS, NUM_CLASS, XYSCALE = utils.load_config(FLAGS)
    input_size = FLAGS.size
    video_path = FLAGS.video

    print("Video from: ", video_path )
    try:
        vid = cv2.VideoCapture(int(video_path))
    except:
        vid = cv2.VideoCapture(video_path)

    if FLAGS.framework == 'tflite':
        interpreter = tf.lite.Interpreter(model_path=FLAGS.weights)
        interpreter.allocate_tensors()
        input_details = interpreter.get_input_details()
        output_details = interpreter.get_output_details()
        print(input_details)
        print(output_details)
    else:
        saved_model_loaded = tf.saved_model.load(FLAGS.weights, tags=[tag_constants.SERVING])
        infer = saved_model_loaded.signatures['serving_default']
    
    if FLAGS.output:
        # by default VideoCapture returns float instead of int
        width = int(vid.get(cv2.CAP_PROP_FRAME_WIDTH))
        height = int(vid.get(cv2.CAP_PROP_FRAME_HEIGHT))
        fps = int(vid.get(cv2.CAP_PROP_FPS))
        codec = cv2.VideoWriter_fourcc(*FLAGS.output_format)
        out = cv2.VideoWriter(FLAGS.output, codec, fps, (width, height))

    frame_id = 0
    while True:
        return_value, frame = vid.read()
        if return_value:
            frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
            image = Image.fromarray(frame)
        else:
            if frame_id == vid.get(cv2.CAP_PROP_FRAME_COUNT):
                print("Video processing complete")
                break
            raise ValueError("No image! Try with another video format")
        
        frame_size = frame.shape[:2]
        image_data = cv2.resize(frame, (input_size, input_size))
        image_data = image_data / 255.
        image_data = image_data[np.newaxis, ...].astype(np.float32)
        prev_time = time.time()

        if FLAGS.framework == 'tflite':
            interpreter.set_tensor(input_details[0]['index'], image_data)
            interpreter.invoke()
            pred = [interpreter.get_tensor(output_details[i]['index']) for i in range(len(output_details))]
            if FLAGS.model == 'yolov3' and FLAGS.tiny == True:
                boxes, pred_conf = filter_boxes(pred[1], pred[0], score_threshold=0.25,
                                                input_shape=tf.constant([input_size, input_size]))
            else:
                boxes, pred_conf = filter_boxes(pred[0], pred[1], score_threshold=0.25,
                                                input_shape=tf.constant([input_size, input_size]))
        else:
            batch_data = tf.constant(image_data)
            pred_bbox = infer(batch_data)
            for key, value in pred_bbox.items():
                boxes = value[:, :, 0:4]
                pred_conf = value[:, :, 4:]

        boxes, scores, classes, valid_detections = tf.image.combined_non_max_suppression(
            boxes=tf.reshape(boxes, (tf.shape(boxes)[0], -1, 1, 4)),
            scores=tf.reshape(
                pred_conf, (tf.shape(pred_conf)[0], -1, tf.shape(pred_conf)[-1])),
            max_output_size_per_class=50,
            max_total_size=50,
            iou_threshold=FLAGS.iou,
            score_threshold=FLAGS.score
        )
        pred_bbox = [boxes.numpy(), scores.numpy(), classes.numpy(), valid_detections.numpy()]
        image = utils.draw_bbox(frame, pred_bbox)
        curr_time = time.time()
        exec_time = curr_time - prev_time
        result = np.asarray(image)
        info = "time: %.2f ms" %(1000*exec_time)
        print(info)

        result = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
        if not FLAGS.dis_cv2_window:
            cv2.namedWindow("result", cv2.WINDOW_AUTOSIZE)
            cv2.imshow("result", result)
            if cv2.waitKey(1) & 0xFF == ord('q'): break

        if FLAGS.output:
            out.write(result)

        frame_id += 1

#Reid main
def main(args=None):
    print('\nExecuting reid function\n')

    FLAGS.batch_size = 1


    learning_rate = tf.compat.v1.placeholder(tf.float32, name='learning_rate')
    images = tf.compat.v1.placeholder(tf.float32, [2, FLAGS.batch_size, IMAGE_HEIGHT, IMAGE_WIDTH, 3], name='images')
    labels = tf.compat.v1.placeholder(tf.float32, [FLAGS.batch_size, 2], name='labels')
    is_train = tf.compat.v1.placeholder(tf.bool, name='is_train')
    global_step = tf.Variable(0, name='global_step', trainable=False)
    weight_decay = 0.0005
    tarin_num_id = 0
    val_num_id = 0

    images1, images2 = preprocess(images, is_train)
    #image3, Testimage = preprocess(images, is_train)

    print('Build network')
    logits = network(images1, images2, weight_decay)
    loss = tf.reduce_mean(input_tensor=tf.nn.softmax_cross_entropy_with_logits(labels=tf.stop_gradient(labels), logits=logits))
    inference = tf.nn.softmax(logits)

    optimizer = tf.compat.v1.train.MomentumOptimizer(learning_rate, momentum=0.9)
    train = optimizer.minimize(loss, global_step=global_step)

    print('\nReid function finish\n')

def reid(img1,img2):
    with tf.compat.v1.Session() as sess:
        sess.run(tf.compat.v1.global_variables_initializer())
        saver = tf.compat.v1.train.Saver()

        ckpt = tf.train.get_checkpoint_state(FLAGS.logs_dir)
        if ckpt and ckpt.model_checkpoint_path:
            print('Restore model')
            saver.restore(sess, ckpt.model_checkpoint_path)

        try:
            image1 = img1
            image1 = cv2.resize(image1, (IMAGE_WIDTH, IMAGE_HEIGHT))
            image1 = cv2.cvtColor(image1, cv2.COLOR_BGR2RGB)
            image1 = np.reshape(image1, (1, IMAGE_HEIGHT, IMAGE_WIDTH, 3)).astype(float)
            image2 = img2
            image2 = cv2.resize(image2, (IMAGE_WIDTH, IMAGE_HEIGHT))
            image2 = cv2.cvtColor(image2, cv2.COLOR_BGR2RGB)
            image2 = np.reshape(image2, (1, IMAGE_HEIGHT, IMAGE_WIDTH, 3)).astype(float)
            test_images = np.array([image1, image2])

            feed_dict = {images: test_images, is_train: False}
            prediction = sess.run(inference, feed_dict=feed_dict)
            return (bool(not np.argmax(prediction[0])))
        except:
            print('Error')
            return False

#Run all program
if __name__ == '__main__':
    try:
        tf.compat.v1.app.run()
        app.run(yolo)
    except SystemExit:
        pass

My problem is that I don't know how to run both tensorflow app at the same time to avoid loading each time I use person-reID to check the identity of each person detected by Yolov4.

Thank you so much

0 Answers
Related