train_cls.py

#  Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
#
#Licensed under the Apache License, Version 2.0 (the "License");
#you may not use this file except in compliance with the License.
#You may obtain a copy of the License at
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
#Unless required by applicable law or agreed to in writing, software
#distributed under the License is distributed on an "AS IS" BASIS,
#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#See the License for the specific language governing permissions and
#limitations under the License.

import os
import sys
import time
import shutil
import argparse
import ast
import logging
import numpy as np
import paddle.fluid as fluid
import paddle.fluid.framework as framework

from models import *
from data.modelnet40_reader import ModelNet40ClsReader
from data.data_utils import *
from utils import *

logging.root.handlers = []
FORMAT = '%(asctime)s-%(levelname)s: %(message)s'
logging.basicConfig(level=logging.INFO, format=FORMAT, stream=sys.stdout)
logger = logging.getLogger(__name__)


def parse_args():
    parser = argparse.ArgumentParser("PointNet++ classification train script")
    parser.add_argument(
        '--model',
        type=str,
        default='MSG',
        help='SSG or MSG model to train, default MSG')
    parser.add_argument(
        '--use_gpu',
        type=ast.literal_eval,
        default=True,
        help='default use gpu.')
    parser.add_argument(
        '--batch_size',
        type=int,
        default=16,
        help='training batch size, default 16')
    parser.add_argument(
        '--num_points',
        type=int,
        default=4096,
        help='number of points in a sample, default: 4096')
    parser.add_argument(
        '--num_classes',
        type=int,
        default=40,
        help='number of classes in dataset, default: 40')
    parser.add_argument(
        '--lr',
        type=float,
        default=0.01,
        help='initial learning rate, default 0.01')
    parser.add_argument(
        '--lr_decay',
        type=float,
        default=0.7,
        help='learning rate decay gamma, default 0.5')
    parser.add_argument(
        '--bn_momentum',
        type=float,
        default=0.99,
        help='initial batch norm momentum, default 0.99')
    parser.add_argument(
        '--decay_steps',
        type=int,
        default=12500,
        help='learning rate and batch norm momentum decay steps, default 12500')
    parser.add_argument(
        '--weight_decay',
        type=float,
        default=1e-5,
        help='L2 regularization weight decay coeff, default 1e-5.')
    parser.add_argument(
        '--epoch',
        type=int,
        default=201,
        help='epoch number. default 201.')
    parser.add_argument(
        '--data_dir',
        type=str,
        default='dataset/ModelNet40/modelnet40_ply_hdf5_2048',
        help='dataset directory')
    parser.add_argument(
        '--save_dir',
        type=str,
        default='checkpoints_cls',
        help='directory name to save train snapshoot')
    parser.add_argument(
        '--resume',
        type=str,
        default=None,
        help='path to resume training based on previous checkpoints. '
        'None for not resuming any checkpoints.')
    parser.add_argument(
        '--log_interval',
        type=int,
        default=1,
        help='mini-batch interval for logging.')
    parser.add_argument(
        '--enable_ce',
        action='store_true',
        help='The flag indicating whether to run the task '
        'for continuous evaluation.')
    args = parser.parse_args()
    return args


def train():
    args = parse_args()
    print_arguments(args)
    # check whether the installed paddle is compiled with GPU
    check_gpu(args.use_gpu)

    if not os.path.isdir(args.save_dir):
        os.makedirs(args.save_dir)

    assert args.model in ['MSG', 'SSG'], \
            "--model can only be 'MSG' or 'SSG'"

    # build model
    if args.enable_ce:
        SEED = 102
        fluid.default_main_program().random_seed = SEED
        framework.default_startup_program().random_seed = SEED

    startup = fluid.Program()
    train_prog = fluid.Program()
    with fluid.program_guard(train_prog, startup):
        with fluid.unique_name.guard():
            train_model = PointNet2ClsMSG(args.num_classes, args.num_points) \
                            if args.model == "MSG" else \
                          PointNet2ClsSSG(args.num_classes, args.num_points)
            train_model.build_model(bn_momentum=args.bn_momentum)
            train_feeds = train_model.get_feeds()
            train_pyreader = train_model.get_pyreader()
            train_outputs = train_model.get_outputs()
            train_loss = train_outputs['loss']
            lr = fluid.layers.exponential_decay(
                    learning_rate=args.lr,
                    decay_steps=args.decay_steps,
                    decay_rate=args.lr_decay,
                    staircase=True)
            lr = fluid.layers.clip(lr, 1e-5, args.lr)
            optimizer = fluid.optimizer.Adam(learning_rate=lr,
                    regularization=fluid.regularizer.L2Decay(args.weight_decay))
            optimizer.minimize(train_loss)
    train_keys, train_values = parse_outputs(train_outputs)

    test_prog = fluid.Program()
    with fluid.program_guard(test_prog, startup):
        with fluid.unique_name.guard():
            test_model = PointNet2ClsMSG(args.num_classes, args.num_points) \
                           if args.model == "MSG" else \
                         PointNet2ClsSSG(args.num_classes, args.num_points)
            test_model.build_model()
            test_feeds = test_model.get_feeds()
            test_outputs = test_model.get_outputs()
            test_pyreader = test_model.get_pyreader()
    test_prog = test_prog.clone(True)
    test_keys, test_values = parse_outputs(test_outputs)

    place = fluid.CUDAPlace(0) if args.use_gpu else fluid.CPUPlace()
    exe = fluid.Executor(place)
    exe.run(startup)

    if args.resume:
        assert os.path.exists(args.resume), \
                "Given resume weight dir {} not exist.".format(args.resume)
        def if_exist(var):
            return os.path.exists(os.path.join(args.resume, var.name))
        fluid.io.load_vars(
            exe, args.resume, predicate=if_exist, main_program=train_prog)

    build_strategy = fluid.BuildStrategy()
    build_strategy.memory_optimize = False
    build_strategy.enable_inplace = False
    build_strategy.fuse_all_optimizer_ops = False
    train_compile_prog = fluid.compiler.CompiledProgram(
            train_prog).with_data_parallel(loss_name=train_loss.name,
                    build_strategy=build_strategy)
    test_compile_prog = fluid.compiler.CompiledProgram(test_prog)

    def save_model(exe, prog, path):
        if os.path.isdir(path):
            shutil.rmtree(path)
        logger.info("Save model to {}".format(path))
        fluid.io.save_persistables(exe, path, prog)

    # get reader
    trans_list = [
        PointcloudScale(),
        PointcloudRotate(),
        PointcloudRotatePerturbation(),
        PointcloudTranslate(),
        PointcloudJitter(),
        PointcloudRandomInputDropout(),
    ]
    modelnet_reader = ModelNet40ClsReader(args.data_dir, mode='train', transforms=trans_list)
    train_reader = modelnet_reader.get_reader(args.batch_size, args.num_points)
    train_pyreader.decorate_sample_list_generator(train_reader, place)
    modelnet_reader = ModelNet40ClsReader(args.data_dir, mode='test', transforms=None)
    test_reader = modelnet_reader.get_reader(args.batch_size, args.num_points)
    test_pyreader.decorate_sample_list_generator(test_reader, place)

    train_stat = Stat()
    test_stat = Stat()

    ce_time = 0
    ce_loss = []

    for epoch_id in range(args.epoch):
        try:
            train_pyreader.start()
            train_iter = 0
            train_periods = []
            while True:
                cur_time = time.time()
                train_outs = exe.run(train_compile_prog, fetch_list=train_values + [lr.name])
                period = time.time() - cur_time
                train_periods.append(period)
                train_stat.update(train_keys, train_outs[:-1])
                if train_iter % args.log_interval == 0:
                    log_str = ""
                    for name, values in zip(train_keys + ['learning_rate'], train_outs):
                        log_str += "{}: {:.5f}, ".format(name, np.mean(values))
                        if name == 'loss':
                            ce_loss.append(np.mean(values))
                    logger.info("[TRAIN] Epoch {}, batch {}: {}time: {:.2f}".format(epoch_id, train_iter, log_str, period))
                train_iter += 1
        except fluid.core.EOFException:
            logger.info("[TRAIN] Epoch {} finished, {}average time: {:.2f}".format(epoch_id, train_stat.get_mean_log(), np.mean(train_periods[1:])))
            ce_time = np.mean(train_periods[1:])
            save_model(exe, train_prog, os.path.join(args.save_dir, str(epoch_id)))

            # evaluation
            if not args.enable_ce:
                try:
                    test_pyreader.start()
                    test_iter = 0
                    test_periods = []
                    while True:
                        cur_time = time.time()
                        test_outs = exe.run(test_compile_prog, fetch_list=test_values)
                        period = time.time() - cur_time
                        test_periods.append(period)
                        test_stat.update(test_keys, test_outs)
                        if test_iter % args.log_interval == 0:
                            log_str = ""
                            for name, value in zip(test_keys, test_outs):
                                log_str += "{}: {:.4f}, ".format(name, np.mean(value))
                            logger.info("[TEST] Epoch {}, batch {}: {}time: {:.2f}".format(epoch_id, test_iter, log_str, period))
                        test_iter += 1
                except fluid.core.EOFException:
                    logger.info("[TEST] Epoch {} finished, {}average time: {:.2f}".format(epoch_id, test_stat.get_mean_log(), np.mean(test_periods[1:])))
                finally:
                    test_pyreader.reset()
                    test_stat.reset()
                    test_periods = []

        finally:
            train_pyreader.reset()
            train_stat.reset()
            train_periods = []

    # only for ce
    if args.enable_ce:
        card_num = get_cards()
        _loss = 0
        _time = 0
        try:
            _time = ce_time
            _loss = np.mean(ce_loss[1:])
        except:
            print("ce info error")
        print("kpis\ttrain_cls_%s_duration_card%s\t%s" % (args.model, card_num, _time))
        print("kpis\ttrain_cls_%s_loss_card%s\t%f" % (args.model, card_num, _loss))

def get_cards():
    num = 0
    cards = os.environ.get('CUDA_VISIBLE_DEVICES', '')
    if cards != '':
        num = len(cards.split(","))
    return num

if __name__ == "__main__":
    train()