# Single class pedestrian detector for SOD RealNets.
#
# This file can serve as template for your future RealNets models to be generated by the SOD training
# interfaces which are documented at https://sod.pixlab.io/api.html#realnet_train.
#
# Copyright (C) PixLab| Symisc Systems - All right reserved. legal@symisc.net - https://sod.pixlab.io

# The first thing to specify is where the training samples are located.
# You must group your dataset on the same directory so can SOD load each entry 
# on a single run and pass the collected image set to the RealNet trainer.

[paths]

# Mandatory positive samples path (i.e. the pedestrian dataset that may contains hundred or thousand of images)
pos = /var/pedestrian_dataset/positives

# Background samples path (i.e. various negative samples holding anything [car, trees, bus, cat, etc] except a pedestrian!! very important)
neg = /var/pedestrian_dataset/background

# Optional test sample path
#test = /var/pedestrian_dataset/test

# True to recurse (scan) subdirectories on the root path of your dataset (positives, background and test paths)
recurse = true

# Everything below is an optional field and does not require that you mess with it unless
# you know what you doing (i.e. Tune your model)

[detector]

# min_tree_depth = 6  # Minimum tree depth

# max_tree_depth = 12 # Maximum tree depth

# max_trees = 2048 # Maximum decision tress to generate for this model

# tpr = 0.9975 # Minimum True Positive Rate (TPR) which must be a float value set between 0.1 .. 1

# fpr = 0.5 # Maximum False Positive Rate (FPR) which must be a float value set between 0.1 .. 1

# data_augment = false # Introduce small perturbation to the input positive samples

# target_fpr = 1e-6 # Target false positive rate (FPR) to achieve. 
                    # When we hit this value or max_trees whichever occurs first, training is stopped.

# normalize = false # Normalize the training positive samples
	
# Information about your model
name = pedestrian
about = RealNets pedestrian detector (single class) - Copyright (C) 2017 - 2018 Symisc Systems


