Darknet21.py

# ==============================================================================
# MIT License
#
# Copyright 2021 Institute for Automotive Engineering of RWTH Aachen University.
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
# ==============================================================================

"""Model configuration for Darknet21"""
import numpy as np
from easydict import EasyDict


def Darknet21():
  mc                    = EasyDict()

  mc.CLASSES            = ['Road',
                           'Sidewalk',
                           'Building',
                           'Pole',
                           'Vegetation',
                           'Person',
                           'TwoWheeler',
                           'Car',
                           'Truck',
                           'Bus',
                           "None"]
  mc.NUM_CLASS          = len(mc.CLASSES)
  mc.CLS_2_ID           = dict(zip(mc.CLASSES, range(len(mc.CLASSES))))
  mc.CLS_LOSS_WEIGHT    = np.array([1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0])
  mc.CLS_COLOR_MAP      = np.array([[128,  64, 128],  # Road
                                    [244,  35, 232],  # Sidewalk
                                    [ 70,  70,  70],  # Building
                                    [153, 153, 153],  # Pole
                                    [107, 142,  35],  # Vegetation
                                    [220,  20,  60],  # Person
                                    [255,   0,   0],  # Two Wheeler
                                    [  0,   0, 142],  # Car
                                    [  0,   0,  70],  # Truck
                                    [  0,  60, 100],  # Bus
                                    [  0,   0,   0]   # None
                                    ]) / 255.0

  # Input Shape
  mc.BATCH_SIZE         = 16
  mc.AZIMUTH_LEVEL      = 240
  mc.ZENITH_LEVEL       = 32
  mc.NUM_FEATURES       = 6

  # Loss
  mc.USE_FOCAL_LOSS     = False  # either use focal loss or sparse categorical cross entropy
  mc.FOCAL_GAMMA        = 2.0
  mc.CLS_LOSS_COEF      = 15.0
  mc.DENOM_EPSILON      = 1e-12   # small value used in denominator to prevent division by 0

  # Gradient Decent
  mc.LEARNING_RATE      = 0.01
  mc.LR_DECAY_STEPS     = 500
  mc.LR_DECAY_FACTOR    = 0.99
  mc.MAX_GRAD_NORM      = 1.0

  # Network
  mc.DROP_RATE          = 0.01
  mc.BN_MOMENTUM        = 0.9
  mc.NUM_LAYERS         = 21
  mc.OUTPUT_STRIDE      = 16  # Output stride only horizontally

  # Dataset
  mc.DATA_AUGMENTATION  = True
  mc.RANDOM_FLIPPING    = True
  mc.SHIFT_UP_DOWN      = 0
  mc.SHIFT_LEFT_RIGHT   = 70

  # x, y, z, intensity, distance
  mc.INPUT_MEAN         = np.array([[[24.810, 0.819, 0.000, 16.303, 25.436]]])
  mc.INPUT_STD          = np.array([[[30.335, 7.807, 2.058, 25.208, 30.897]]])

  return mc