training_3DMatch.py

# Common libs
import time
import os
import sys

# Custom libs
from utils.config import Config
from utils.trainer import ModelTrainer
from models.KPFCNN_model import KernelPointFCNN

# Dataset
from datasets.ThreeDMatch import ThreeDMatchDataset


# ----------------------------------------------------------------------------------------------------------------------
#
#           Config Class
#       \******************/
#


class ThreeDMatchConfig(Config):
    """
    Override the parameters you want to modify for this dataset
    """

    ####################
    # Dataset parameters
    ####################
    is_test = False
    gpu_id = 0
    dataset = '3DMatch'

    # Number of CPU threads for the input pipeline
    input_threads = 8

    #########################
    # Architecture definition
    #########################

    architecture = ['simple',
                    'resnetb',
                    'resnetb_strided',
                    'resnetb',
                    'resnetb_strided',
                    'resnetb', 
                    'resnetb_strided', 
                    'resnetb', 
                    'resnetb_strided', 
                    'resnetb',
                    'nearest_upsample',
                    'unary',
                    'nearest_upsample',
                    'unary',
                    'nearest_upsample',
                    'unary',
                    'nearest_upsample',
                    'unary',
                    'last_unary']

    # KPConv specific parameters
    num_kernel_points = 15
    first_subsampling_dl = 0.03

    # Density of neighborhoods for deformable convs (which need bigger radiuses). For normal conv we use KP_extent
    density_parameter = 5.0

    # Influence function of KPConv in ('constant', 'linear', gaussian)
    KP_influence = 'linear'
    KP_extent = 1.0

    # Aggregation function of KPConv in ('closest', 'sum')
    convolution_mode = 'sum'

    # Can the network learn modulations in addition to deformations
    modulated = False

    # detector loss weight
    det_loss_weight = 1

    # Offset loss
    # 'permissive' only constrains offsets inside the big radius
    # 'fitting' helps deformed kernels to adapt to the geometry by penalizing distance to input points
    offsets_loss = 'fitting'
    offsets_decay = 0.1

    # Choice of input features
    in_features_dim = 1

    # Batch normalization parameters
    use_batch_norm = True
    batch_norm_momentum = 0.98

    # batch hard loss safe radius
    safe_radius = 0.1

    #####################
    # Training parameters
    #####################

    # Maximal number of epochs
    max_epoch = 200

    # Learning rate management
    learning_rate = 1e-1
    momentum = 0.98
    lr_decays = {i: 0.1 ** (1 / 80) for i in range(1, max_epoch)}
    grad_clip_norm = 100.0

    # Number of batch
    batch_num = 1
    # Number of keypoints
    keypts_num = 256

    # Number of steps per epochs (cannot be None for this dataset)
    epoch_steps = 5000

    # Number of validation examples per epoch
    validation_size = 500

    # Number of epoch between each snapshot
    snapshot_gap = 1

    # Augmentations
    augment_scale_anisotropic = True
    augment_symmetries = [False, False, False]
    augment_rotation = 1
    augment_scale_min = 0.9
    augment_scale_max = 1.1
    augment_noise = 0.005
    augment_occlusion = 'none'

    # Do we nee to save convergence
    saving = True
    saving_path = None


# ----------------------------------------------------------------------------------------------------------------------
#
#           Main Call
#       \***************/
#


if __name__ == '__main__':
    ##########################
    # Initiate the environment
    ##########################

    # Enable/Disable warnings (set level to '0'/'3')
    os.environ['TF_CPP_MIN_LOG_LEVEL'] = '0'

    ###########################
    # Load the model parameters
    ###########################

    config = ThreeDMatchConfig()

    ##############
    # Prepare Data
    ##############

    print()
    print('Dataset Preparation')
    print('*******************')

    # Initiate dataset configuration
    dataset = ThreeDMatchDataset(config.input_threads, voxel_size=config.first_subsampling_dl)

    # Create subsampled input clouds
    dl0 = config.first_subsampling_dl
    # dataset.load_subsampled_clouds(dl0)

    # Initialize input pipelines
    dataset.init_input_pipeline(config)

    # Test the input pipeline alone with this debug function
    # dataset.check_input_pipeline_timing(config)

    ##############
    # Define Model
    ##############

    print('Creating Model')
    print('**************\n')
    t1 = time.time()

    # Model class
    model = KernelPointFCNN(dataset.flat_inputs, config)

    # Trainer class
    trainer = ModelTrainer(model)
    # trainer = ModelTrainer(model, restore_snap='results/Log_/snapshots/snap-')
    t2 = time.time()

    print('\n----------------')
    print('Done in {:.1f} s'.format(t2 - t1))
    print('----------------\n')

    ################
    # Start training
    ################

    print('Start Training')
    print('**************\n')

    trainer.train(model, dataset)