train_contrastnet.py

import argparse
import math
import h5py
import numpy as np
import tensorflow as tf
import socket
import importlib
import os
import sys
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.append(BASE_DIR)
sys.path.append(os.path.join(BASE_DIR, 'models'))
sys.path.append(os.path.join(BASE_DIR, 'utils'))
import provider
import tf_util

parser = argparse.ArgumentParser()
parser.add_argument('--gpu', type=int, default=0, help='GPU to use [default: GPU 0]')
parser.add_argument('--model', default='contrastnet', help='Model name: contrastnet')
parser.add_argument('--log_dir', default='log', help='Log dir [default: log]')
parser.add_argument('--num_point', type=int, default=512, help='Point Number [256/512/1024/2048] [default: 1024]')
parser.add_argument('--max_epoch', type=int, default=600, help='Epoch to run [default: 250]')
parser.add_argument('--batch_size', type=int, default=40, help='Batch Size during training [default: 32]')
parser.add_argument('--learning_rate', type=float, default=0.001, help='Initial learning rate [default: 0.001]')
parser.add_argument('--momentum', type=float, default=0.9, help='Initial learning rate [default: 0.9]')
parser.add_argument('--optimizer', default='adam', help='adam or momentum [default: adam]')
parser.add_argument('--decay_step', type=int, default=200000, help='Decay step for lr decay [default: 200000]')
parser.add_argument('--decay_rate', type=float, default=0.7, help='Decay rate for lr decay [default: 0.8]')
parser.add_argument('--model_path', default='log/model.ckpt', help='model checkpoint file path [default: log/model.ckpt]')

FLAGS = parser.parse_args()


BATCH_SIZE = FLAGS.batch_size
NUM_POINT = FLAGS.num_point
MAX_EPOCH = FLAGS.max_epoch
BASE_LEARNING_RATE = FLAGS.learning_rate
GPU_INDEX = FLAGS.gpu
MOMENTUM = FLAGS.momentum
OPTIMIZER = FLAGS.optimizer
DECAY_STEP = FLAGS.decay_step
DECAY_RATE = FLAGS.decay_rate
MODEL_PATH = FLAGS.model_path

MODEL = importlib.import_module(FLAGS.model) # import network module
MODEL_FILE = os.path.join(BASE_DIR, 'models', FLAGS.model+'.py')
LOG_DIR = FLAGS.log_dir
if not os.path.exists(LOG_DIR): os.mkdir(LOG_DIR)
os.system('cp %s %s' % (MODEL_FILE, LOG_DIR)) # bkp of model def
os.system('cp train.py %s' % (LOG_DIR)) # bkp of train procedure
LOG_FOUT = open(os.path.join(LOG_DIR, 'log_train.txt'), 'w')
LOG_FOUT.write(str(FLAGS)+'\n')

MAX_NUM_POINT = 2048
NUM_CLASSES = 2

BN_INIT_DECAY = 0.5
BN_DECAY_DECAY_RATE = 0.5
BN_DECAY_DECAY_STEP = float(DECAY_STEP)
BN_DECAY_CLIP = 0.99

HOSTNAME = socket.gethostname()

# ModelNet40 official train/test split
TRAIN_FILES = provider.getDataFiles( \
    os.path.join(BASE_DIR, 'data/modelnet40_ply_hdf5_2048_cut/train_files.txt'))
TEST_FILES = provider.getDataFiles(\
    os.path.join(BASE_DIR, 'data/modelnet40_ply_hdf5_2048_cut/test_files.txt'))

def log_string(out_str):
    LOG_FOUT.write(out_str+'\n')
    LOG_FOUT.flush()
    print(out_str)


def get_learning_rate(batch):
    learning_rate = tf.train.exponential_decay(
                        BASE_LEARNING_RATE,  # Base learning rate.
                        batch * BATCH_SIZE,  # Current index into the dataset.
                        DECAY_STEP,          # Decay step.
                        DECAY_RATE,          # Decay rate.
                        staircase=True)
    learning_rate = tf.maximum(learning_rate, 0.00001) # CLIP THE LEARNING RATE!
    return learning_rate

def get_bn_decay(batch):
    bn_momentum = tf.train.exponential_decay(
                      BN_INIT_DECAY,
                      batch*BATCH_SIZE,
                      BN_DECAY_DECAY_STEP,
                      BN_DECAY_DECAY_RATE,
                      staircase=True)
    bn_decay = tf.minimum(BN_DECAY_CLIP, 1 - bn_momentum)
    return bn_decay

def train():
    with tf.Graph().as_default():
        with tf.device('/gpu:'+str(GPU_INDEX)):
            pointclouds_pl_1, labels_pl = MODEL.placeholder_inputs(BATCH_SIZE, NUM_POINT)
            pointclouds_pl_2, labels_pl = MODEL.placeholder_inputs(BATCH_SIZE, NUM_POINT)
            is_training_pl = tf.placeholder(tf.bool, shape=())
            print(is_training_pl)

            # Note the global_step=batch parameter to minimize.
            # That tells the optimizer to helpfully increment the 'batch' parameter for you every time it trains.
            batch = tf.Variable(0)
            bn_decay = get_bn_decay(batch)
            tf.summary.scalar('bn_decay', bn_decay)

            # Get model and loss
            pred, feature1, feature2, end_points = MODEL.get_model(pointclouds_pl_1, pointclouds_pl_2, is_training_pl, bn_decay=bn_decay)
            loss = MODEL.get_loss(pred, labels_pl, end_points)
            tf.summary.scalar('loss', loss)

            correct = tf.equal(tf.argmax(pred, 1), tf.to_int64(labels_pl))
            accuracy = tf.reduce_sum(tf.cast(correct, tf.float32)) / float(BATCH_SIZE)
            tf.summary.scalar('accuracy', accuracy)

            # Get training operator
            learning_rate = get_learning_rate(batch)
            tf.summary.scalar('learning_rate', learning_rate)
            if OPTIMIZER == 'momentum':
                optimizer = tf.train.MomentumOptimizer(learning_rate, momentum=MOMENTUM)
            elif OPTIMIZER == 'adam':
                optimizer = tf.train.AdamOptimizer(learning_rate)
            train_op = optimizer.minimize(loss, global_step=batch)

            # Add ops to save and restore all the variables.
            saver = tf.train.Saver(max_to_keep = 10)

        # Create a session
        config = tf.ConfigProto()
        config.gpu_options.allow_growth = True
        config.allow_soft_placement = True
        config.log_device_placement = False
        sess = tf.Session(config=config)

        # Add summary writers
        #merged = tf.merge_all_summaries()
        merged = tf.summary.merge_all()
        train_writer = tf.summary.FileWriter(os.path.join(LOG_DIR, 'train'),
                                  sess.graph)
        test_writer = tf.summary.FileWriter(os.path.join(LOG_DIR, 'test'))

        # Init variables
        init = tf.global_variables_initializer()

        sess.run(init, {is_training_pl: True})

        ops = {'pointclouds_pl_1': pointclouds_pl_1,
               'pointclouds_pl_2': pointclouds_pl_2,
               'labels_pl': labels_pl,
               'is_training_pl': is_training_pl,
               'pred': pred,
               'loss': loss,
               'train_op': train_op,
               'merged': merged,
               'step': batch}

        for epoch in range(MAX_EPOCH):
            log_string('**** EPOCH %03d ****' % (epoch))
            sys.stdout.flush()

            train_one_epoch(sess, ops, train_writer)

            if epoch % 40 == 0 and epoch >= 120:
                save_path = saver.save(sess, os.path.join(LOG_DIR, 'epoch_' + str(epoch)+'.ckpt'))
                log_string("Model saved in file: %s" % save_path)
            elif epoch % 10 == 0:
                save_path = saver.save(sess, os.path.join(LOG_DIR, 'model.ckpt'))
                log_string("Model saved in file: %s" % save_path)


def train_one_epoch(sess, ops, train_writer):
    """ ops: dict mapping from string to tf ops """
    is_training = True

    # Shuffle train files
    train_file_idxs = np.arange(0, len(TRAIN_FILES))
    np.random.shuffle(train_file_idxs)

    current_data_1 = np.empty([3*len(TRAIN_FILES), NUM_POINT, 3], dtype=float)
    current_data_2 = np.empty([3*len(TRAIN_FILES), NUM_POINT, 3], dtype=float)
    current_label  =  np.empty([3*len(TRAIN_FILES),1], dtype=int)

    fn = 0
    count = 0
    while fn < len(TRAIN_FILES) - 1:
        # log_string('----' + str(fn) + '-----')

        total_current = [];
        a1, a2, _ = provider.loadDataFile_cut_2(TRAIN_FILES[train_file_idxs[fn]])

        idx = np.random.randint(a1.shape[0], size=NUM_POINT)
        a1 = a1[idx,:]
        idx = np.random.randint(a2.shape[0], size=NUM_POINT)
        a2 = a2[idx,:]
        total_current.append(a1)
        total_current.append(a2)

        fn = fn + 1;

        b1, b2, _ = provider.loadDataFile_cut_2(TRAIN_FILES[train_file_idxs[fn]])

        idx = np.random.randint(b1.shape[0], size=NUM_POINT)
        b1 = b1[idx,:]
        idx = np.random.randint(b2.shape[0], size=NUM_POINT)
        b2 = b2[idx,:]
        total_current.append(b1)
        total_current.append(b2)

        fn = fn + 1;

        pair_num = 0
        for index in range(len(total_current)):
            for index2 in range(index + 1, len(total_current)):
                current_data_1[6*count+pair_num,:,:] = total_current[index]
                current_data_2[6*count+pair_num, :,:] = total_current[index2]
                if (index < 2) and (index2 >= 2):
                    current_label[6*count+pair_num,:] = 0
                else:
                    current_label[6*count+pair_num,:] = 1

                pair_num = pair_num + 1
        count = count + 1

    current_label = np.squeeze(current_label)

    file_size = current_data_1.shape[0]
    num_batches = file_size // BATCH_SIZE

    total_correct = 0
    total_seen = 0
    loss_sum = 0

    for batch_idx in range(num_batches):
        start_idx = batch_idx * BATCH_SIZE
        end_idx = (batch_idx+1) * BATCH_SIZE

        # shuffle each batch
        data_1 = current_data_1[start_idx:end_idx, :, :]
        data_2 = current_data_2[start_idx:end_idx, :, :]
        label = current_label[start_idx:end_idx]
        combine_data = np.concatenate((data_1, data_2), axis=2)
        combine_data, label, _ = provider.shuffle_data(combine_data, np.squeeze(label))
        data_1 = combine_data[:, :, 0:3]
        data_2 = combine_data[:, :, 3:6]
        label = np.squeeze(label)

        # Augment batched point clouds by rotation and jittering
        rotated_data_1 = provider.rotate_point_cloud(data_1)
        jittered_data_1 = provider.jitter_point_cloud(rotated_data_1)
        jittered_data_1 = provider.random_scale_point_cloud(jittered_data_1)
        jittered_data_1 = provider.rotate_perturbation_point_cloud(jittered_data_1)
        jittered_data_1 = provider.shift_point_cloud(jittered_data_1)

        rotated_data_2 = provider.rotate_point_cloud(data_2)
        jittered_data_2 = provider.jitter_point_cloud(rotated_data_2)
        jittered_data_2 = provider.random_scale_point_cloud(jittered_data_2)
        jittered_data_2 = provider.rotate_perturbation_point_cloud(jittered_data_2)
        jittered_data_2 = provider.shift_point_cloud(jittered_data_2)

        feed_dict = {ops['pointclouds_pl_1']: jittered_data_1,
                     ops['pointclouds_pl_2']: jittered_data_2,
                     ops['labels_pl']: label,
                     ops['is_training_pl']: is_training}
        summary, step, _, loss_val, pred_val = sess.run([ops['merged'], ops['step'],
            ops['train_op'], ops['loss'], ops['pred']], feed_dict=feed_dict)
        train_writer.add_summary(summary, step)
        pred_val = np.argmax(pred_val, 1)
        correct = np.sum(pred_val == label)
        total_correct += correct
        total_seen += BATCH_SIZE
        loss_sum += loss_val

        if batch_idx % 50 == 0:
            log_string('mean loss: {0:f}     accuracy: {1:f}'.format(loss_sum / float(batch_idx+1), total_correct / float(total_seen)))

    log_string('mean loss: {0:f}     accuracy: {1:f}'.format(loss_sum / float(batch_idx+1), total_correct / float(total_seen)))

if __name__ == "__main__":
    train()
    LOG_FOUT.close()