Building Head/Face Detection System <2. Code>

Posted On June 20, 2020

This post contains the implementation of the concepts described in previous post in https://alvissalim.com/2020/05/15/building-a-face-detector-system/ Previously, we have defined the expectation of what a head/face detection neural network model may look like. In this post, we will go through the steps of implementing the model in PyTorch. It is assumed that you are already familiar with the PyTorch framework and Python.

The Neural Network Model

"""Object Detection Model 
    @author : Muhammad Sakti Alvissalim (alvissalim@gmail.com)
    Copyright (C) 2020  Muhammad Sakti Alvissalim 
    This program is free software: you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
    the Free Software Foundation, either version 3 of the License, or
    (at your option) any later version.
    This program is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU General Public License for more details.
    You should have received a copy of the GNU General Public License
    along with this program.  If not, see <https://www.gnu.org/licenses/>.
"""


import torch
from torch import nn
import torchvision.models as models

class ObjectDetectionModel(nn.Module):
    """ An object detection model for PyTorch """   
    def __init__(self, n_classes: int = 1, output_grid_size : int = 7, boxes_per_cell : int = 4, **kwarg):
        super(ObjectDetectionModel, self).__init__(**kwarg)
        self.n_classes = n_classes
        self.output_grid_size = output_grid_size
        self.boxes_per_cell = boxes_per_cell
        self.n_cells = output_grid_size * output_grid_size

        boxes_tensor_size = (boxes_per_cell * 5) # objectness score + (x,y,w,h)
        conditional_class_prob_tensor_size = n_classes

        cell_tensor_size = boxes_tensor_size + conditional_class_prob_tensor_size
        self.cell_tensor_size = cell_tensor_size
        base_model = models.mobilenet_v2()

        self.features_length = base_model.last_channel
        self.base_features = base_model.features
        
        self.final_feat = nn.Conv2d(base_model.last_channel, cell_tensor_size, 1, stride=1)

    def forward(self, x):
        x = self.base_features(x)
        x = self.final_feat(x)
        x = nn.functional.adaptive_avg_pool2d(x, self.output_grid_size )#
        x = x.permute(0,2,3,1)
        x = x.flatten(start_dim=1)

        return x

Data

Training