Skip to content

Instantly share code, notes, and snippets.

@PrimadonnaGit

PrimadonnaGit/ssd detector_datasets.py

Created September 3, 2019 11:58
Show Gist options
  • Save PrimadonnaGit/ed3c9f0cd2eb1171dca52f862c85de69 to your computer and use it in GitHub Desktop.
Save PrimadonnaGit/ed3c9f0cd2eb1171dca52f862c85de69 to your computer and use it in GitHub Desktop.
Download ZIP
SSD detector for pytorch
Raw
ssd detector_datasets.py
import torch
from torchvision.utils.data import Dataset
import json
import os
from PIL import Image
from utils import transform
class PascalVOCDataset(Dataset):
"""
A Pytorch Dataset class to be used in a PyTorch DataLoader to create batches.
"""
def __init__(self, data_folder, split, keep_difficult=False):
"""
:param data_folder: folder where data files are stored
:param split: split, on of 'TRAIN' or 'TEST'
:param keep_difficult: keep or discard objects that are considered difficult to detect
"""
self.split = split.upper()
assert self.split in {'TRAIN', 'TEST'}
self.data_folder = data_folder
self.keep_difficult = keep_difficult
# Read data files
with open(os.path.join(data_folder, self.split + '_images.json'), 'r') as j:
self.images = json.load(j)
with open(os.path.join(data_folder, self.split + '_objects.json'), 'r') as j:
self.objects = json.load(j)
assert len(self.images) == len(self.objects)
def __getitem__(self, i):
# Read image
image = Image.open(self.images[i], mode='r')
image = Image.convert('RGB')
# Read objects in this image (bounding boxes, labels, difficulties)
objects = self.objects[i]
boxes = torch.FloatTensor(objects['boxes']) # (n_objects, 4)
labels = torch.LongTensor(objects['labels']) # (n_objects)
difficulties = torch.ByteTensor(objects['difficulties']) # (n_objects)
# Discard difficult objects, if desired
if not self.keep_difficult:
boxes = boxes[1 - difficulties]
labels = labels[1 - difficulties]
difficulties = difficulties[1 - difficulties]
# Apply transformations
image, boxes, labels, difficulties = transform(image, boxes, labels, difficulties, split=self.split)
return image, boxes, labels, difficulties
def __len__(self):
return len(self.images)
def collate_fn(self, batch):
"""
Since each image may have a different number of objects, we need a collate function (to be passed to the DataLoader)
This describes how to combine these tensors of different sizes, We use lists.
Note: this need not be defined in this Class, can be standalone
:param batch: an iterable of N sets from __getitem__()
:return: a tensor of images, lists of varying-size tensors of bounding boxes, labels, and difficulties
"""
images = list()
boxes = list()
labels = list()
difficulties = list()
for b in batch:
images.append(b[0])
boxes.append(b[1])
labels.append(b[2])
difficulties.append(b[3])
images = torch.stack(images, dim=0)
return images, boxes, labels, difficulties # tensor (N, 3, 300, 300), 3 lists of N tensors each
Raw
ssd detector_model.py
from torch import nn
from .utils import *
import torch.nn.functional as F
from math import sqrt
from itertools import product as product
import torchvision
import torch
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
class VGGBase(nn.Module):
"""
VGG base convolutions to produce lower-level feature maps.
"""
def __init__(self):
super(VGGBase, self).__init__()
# Standard convolutional layers in VGG16
self.conv1_1 = nn.Conv2d(3, 64, kernel_size=3, padding=1) # stride = 1, by default
self.conv1_2 = nn.Conv2d(64, 64, kernel_size=3, padding=1)
self.pool1 = nn.MaxPool2d(kernel_size=2, stride=2)
self.conv2_1 = nn.Conv2d(64, 128, kernel_size=3, padding=1)
self.conv2_2 = nn.Conv2d(128, 128, kernel_size=3, padding=1)
self.pool2 = nn.MaxPool2d(kernel_size=2, stride=2)
self.conv3_1 = nn.Conv2d(128, 256, kernel_size=3, padding=1)
self.conv3_2 = nn.Conv2d(256, 256, kernel_size=3, padding=1)
self.conv3_3 = nn.Conv2d(256, 256, kernel_size=3, padding=1)
self.pool3 = nn.MaxPool2d(kernel_size=2, stride=2, ceil_mode=True) # ceiling (not floor) here for even dims
self.conv4_1 = nn.Conv2d(256, 512, kernel_size=3, padding=1)
self.conv4_2 = nn.Conv2d(512, 512, kernel_size=3, padding=1)
self.conv4_3 = nn.Conv2d(512, 512, kernel_size=3, padding=1)
self.pool4 = nn.MaxPool2d(kernel_size=2, stride=2)
self.conv5_1 = nn.Conv2d(512, 512, kernel_size=3, padding=1)
self.conv5_2 = nn.Conv2d(512, 512, kernel_size=3, padding=1)
self.conv5_3 = nn.Conv2d(512, 512, kernel_size=3, padding=1)
self.pool5 = nn.MaxPool2d(kernel_size=3, stride=1, padding=1) # retains size because stride is 1 (and padding)
# Replacements for FC6 and FC7 in VGG16
self.conv6 = nn.Conv2d(512, 1024, kernel_size=3, padding=6, dilation=6) # atrous convolution
self.conv7 = nn.Conv2d(1024, 1024, kernel_size=1)
# Load pretrained layers
self.load_pretrained_layers()
def forward(self, image):
"""
Forward propagation.
:param image: image, a tensor of dimensions (N, 3, 300, 300)
:return: lower-level feature maps conv4_3 and conv7
"""
out = F.relu(self.conv1_1(image)) # (N, 64, 300, 300)
out = F.relu(self.conv1_2(out)) # (N, 64, 300, 300)
out = self.pool1(out) # (N, 64, 150, 150)
out = F.relu(self.conv2_1(out)) # (N, 128, 150, 150)
out = F.relu(self.conv2_2(out)) # (N, 128, 150, 150)
out = self.pool2(out) # (N, 128, 75, 75)
out = F.relu(self.conv3_1(out)) # (N, 256, 75, 75)
out = F.relu(self.conv3_2(out)) # (N, 256, 75, 75)
out = F.relu((self.conv3_3(out))) # (N, 256, 75, 75)
out = self.pool3(out) # (N, 256, 38, 38), it would have been 37 if not for ceil_mode True
out = F.relu(self.conv4_1(out)) # (N, 512, 38, 38)
out = F.relu(self.conv4_2(out)) # (N, 512, 38, 38)
out = F.relu(self.conv4_3(out)) # (N, 512, 38, 38)
conv4_3_feats = out # (N, 512, 38, 38)
out = self.pool4(out) # (N, 512, 19, 19)
out = F.relu(self.conv5_1(out)) # (N, 512, 19, 19)
out = F.relu(self.conv5_2(out)) # (N, 512, 19, 19)
out = F.relu(self.conv5_3(out)) # (N, 512, 19, 19)
out = self.pool5(out) # (N, 512, 19, 19), pool5 does not reduce dimensions
out = F.relu(self.conv6(out)) # (N, 1024, 19, 19)
conv7_feats = F.relu(self.conv7(out)) # (N, 1024, 19, 19)
# Lower-level feature maps
return conv4_3_feats, conv7_feats
def load_pretrained_layers(self):
"""
As in the paper, we use a VGG-16 pretrained on the ImageNet task as the base network.
We copy these parameters into our network. It's straightforward for conv1 to conv5
However, the original VGG-16 does not contain the conv6 and con7 layers.
Therefore, we convert fc6 and fc7 into convolutional layers, and subsample by decimation.
"""
# Current state of base
state_dict =self.state_dict()
param_names = list(state_dict.keys())
# Pretrained VGG base
pretrained_state_dict = torchvision.models.vgg16(pretrained=True).state_dict()
pretrained_param_names = list(pretrained_state_dict.keys())
# Transfer conv. parameters from pretrained model to current model
for i, param in enumerate(param_names[:-4]): # excluding conv6 and conv7 parameters
state_dict[param] = pretrained_state_dict[pretrained_param_names[i]]
# Convert fc6, fc7 to convolutional layers, and subsample (by decimation) to sizes of conv6 and con7
# fc6
conv_fc6_weight = pretrained_state_dict['classifier.0.weight'].view(4096, 512, 7, 7) # (4096, 512, 7, 7)
conv_fc6_bias = pretrained_state_dict['classifier.0.bias'] # (4096)
state_dict['conv6.weight'] = decimate(conv_fc6_weight, m=[4, None, 3, 3]) # (1024, 512, 3, 3)
state_dict['conv6.bias'] = decimate(conv_fc6_bias, m=[4]) # (1024)
# fc7
conv_fc7_weight = pretrained_state_dict['classifier.3.weight'].view(4096, 4096, 1, 1) # (4096, 4096, 1, 1)
conv_fc7_bias = pretrained_state_dict['classifier.3.bias'] # (4096)
state_dict['conv7.weight'] = decimate(conv_fc7_weight, m=[4, 4, None, None]) # (1024, 1024, 1, 1)
state_dict['conv7.bias'] = decimate(conv_fc7_bias, m=[4]) # (1024)
# Note: an FC layer of size (K) opertaion on a flattened version (C*H*W) of a 2D image of size (C, H, W)
# is equivalent to a convolutional layer with kernel size (H, W), input channels C, output channels K
# operating on the 2D image of size (C, H, W) without padding
self.load_state_dict(state_dict)
print("\nLoaded base model.\n")
class AuxiliaryConvolutions(nn.Module):
"""
Additional convolutions to produce higher-level feature maps.
"""
def __init__(self):
super(AuxiliaryConvolutions, self).__init__()
# Auxiliary/additional convolutions on top of the VGG base
self.conv8_1 = nn.Conv2d(1024, 256, kernel_size=1, padding=0) # stride = 1, by default
self.conv8_2 = nn.Conv2d(256, 512, kernel_size=3, stride=2, padding=1) # dim. reduction because stride > 1
self.conv9_1 = nn.Conv2d(512, 128, kernel_size=1, padding=0)
self.conv9_2 = nn.Conv2d(128, 256, kernel_size=3, stride=2, padding=1) # dim. reduction because stride > 1
self.conv10_1 = nn.Conv2d(256, 128, kernel_size=1, padding=0)
self.conv10_2 = nn.Conv2d(128, 256, kernel_size=3, padding=0) # dim. reduction because padding = 0
self.conv11_1 = nn.Conv2d(256, 128, kernel_size=1, padding=0)
self.conv11_2 = nn.Conv2d(128, 256, kernel_size=3, padding=0) # dim. reduction because padding = 0
# Initialize convolutions parameters
self.init_conv2d()
def init_conv2d(self):
"""
Initialize convolution parameters.
"""
for c in self.children():
if isinstance(c, nn.Conv2d):
nn.init.xavier_uniform_(c.weight)
nn.init.constant_(c.bias, 0.)
def forward(self, conv7_feats):
"""
Forward propagation.
:param conv7_feats: lower-level conv7 feature map, a tensor of dimensions (N, 1024, 19, 19)
:return: higher-level feature maps conv8_2, conv9_2, conv10_2, and conv11_2
"""
out = F.relu(self.conv8_1(conv7_feats)) # (N, 256, 19, 19)
out = F.relu(self.conv8_2(out)) # (N, 512, 10, 10)
conv8_2_feats = out # (N, 512, 10, 10)
out = F.relu(self.conv9_1(out)) # (N, 128, 10, 10)
out = F.relu(self.conv9_2(out)) # (N, 256, 5, 5)
conv9_2_feats = out # (N, 256, 5, 5)
out = F.relu(self.conv10_1(out)) # (N, 128, 5, 5)
out = F.relu(self.conv10_2(out)) # (N, 256, 3, 3)
conv10_2_feats = out # (N, 256, 3, 3)
out = F.relu(self.conv11_1(out)) # (N, 128, 3, 3)
conv11_2_feats = F.relu(self.conv11_2(out)) # (N, 256, 1, 1)
# Higher-level feature maps
return conv8_2_feats, conv9_2_feats, conv10_2_feats, conv11_2_feats
class PredictionConvolutions(nn.Module):
"""
Convolutions to predict class scores and bounding boxes using lower and higher-level feature maps.
The bounding boxes (locations) are predicted as encoded offsets w.r.t each of the 8732 prior (default) boxes.
The class scores represent the scores of each object class in each of the 8732 bounding boxes located.
A high score for 'background' = no object.
"""
def __init__(self, n_classes):
"""
:param n_classes: number of different types of objects
"""
super(PredictionConvolutions, self).__init__()
self.n_classes = n_classes
# Number of prior-boxes we are considering per position in each feature map
n_boxes = {'conv4_3': 4,
'conv7' : 6,
'conv8_2': 6,
'conv9_2': 6,
'conv10_2': 4,
'conv11_2': 4}
# 4 prior-boxes implies we use 4 different aspect ratios, etc.
# Localizaition prediction convolutions (predict offsets(4) w.r.t prior-boxes)
self.loc_conv4_3 = nn.Conv2d(512, n_boxes['conv4_3'] * 4, kernel_size=3, padding=1)
self.loc_conv7 = nn.Conv2d(1024, n_boxes['conv7'] * 4, kernel_size=3, padding=1)
self.loc_conv8_2 = nn.Conv2d(512, n_boxes['conv8_2'] * 4, kernel_size=3, padding=1)
self.loc_conv9_2 = nn.Conv2d(256, n_boxes['conv9_2'] * 4, kernel_size=3, padding=1)
self.loc_conv10_2 = nn.Conv2d(256, n_boxes['conv10_2'] * 4, kernel_size=3, padding=1)
self.loc_conv11_2 = nn.Conv2d(256, n_boxes['conv11_2'] * 4, kernel_size=3, padding=1)
# Class prediction convolutions (predict classes in localization boxes)
self.cl_conv4_3 = nn.Conv2d(512, n_boxes['conv4_3'] * n_classes, kernel_size=3, padding=1)
self.cl_conv7 = nn.Conv2d(1024, n_boxes['conv7'] * n_classes, kernel_size=3, padding=1)
self.cl_conv8_2 = nn.Conv2d(512, n_boxes['conv8_2'] * n_classes, kernel_size=3, padding=1)
self.cl_conv9_2 = nn.Conv2d(256, n_boxes['conv9_2'] * n_classes, kernel_size=3, padding=1)
self.cl_conv10_2 = nn.Conv2d(256, n_boxes['conv10_2'] * n_classes, kernel_size=3, padding=1)
self.cl_conv11_2 = nn.Conv2d(256, n_boxes['conv11_2'] * n_classes, kernel_size=3, padding=1)
# Initialize convolutions parameters
self.init_conv2d()
def init_conv2d(self):
"""
Initialize convolution parameters.
"""
for c in self.children():
if isinstance(c, nn.Conv2d):
nn.init.xavier_uniform_(c.weight)
nn.init.constant_(c.bias, 0.)
def forward(self, conv4_3_feats, conv7_feats, conv8_2_feats, conv9_2_feats, conv10_2_feats, conv11_2_feats):
"""
Forward propagation.
:param conv4_3_feats: conv4_3 feature map, a tensor of dimensions (N, 512, 38, 38)
:param conv7_feats: conv7 feature map, a tensor of dimensions (N, 1024, 19, 19)
:param conv8_2_feats: conv8_2 feature map, a tensor of dimensions (N, 512, 10, 10)
:param conv9_2_feats: conv9_2 feature map, a tensor of dimensions (N, 256, 5, 5)
:param conv10_2_feats: conv10_2 feature map, a tensor of dimensions (N, 256, 3, 3)
:param conv11_2_feats: conv11_2 feature map, a tensor of dimensions (N, 256, 1, 1)
:return: 8732 locations and class scores (i.e. w.r.t each prior box) for each image
"""
batch_size = conv4_3_feats.size(0)
# Predict localization boxes bounds (as offset w.r.t prior-boxes)
l_conv4_3 = self.loc_conv4_3(conv4_3_feats) # (N, 16, 38, 38)
l_conv4_3 = l_conv4_3.permute(0, 2, 3,
1).contiguous() # (N, 38, 38, 16), to match prior-box order (after .view())
# (.contiguous() ensures it is stored in a contiguous chunk of memory, needed for .view() below)
l_conv4_3 = l_conv4_3.view(batch_size, -1, 4) # (N, 5776, 4), there are a total 5776 boxes on this feature map
l_conv7 = self.loc_conv7(conv7_feats) # (N, 24, 19, 19)
l_conv7 = l_conv7.permute(0, 2, 3, 1).contiguous() # (N, 19, 19, 24)
l_conv7 = l_conv7.view(batch_size, -1, 4) # (N, 2166, 4), there are total 2166 boxes on this feature map
l_conv8_2 = self.loc_conv8_2(conv8_2_feats) # (N, 24, 10, 10)
l_conv8_2 = l_conv8_2.permute(0, 2, 3, 1).contiguous() # (N, 10, 10, 24)
l_conv8_2 = l_conv8_2.view(batch_size, -1, 4) # (N, 600, 4)
l_conv9_2 = self.loc_conv9_2(conv9_2_feats) # (N, 24, 5, 5)
l_conv9_2 = l_conv9_2.permute(0, 2, 3, 1).contiguous() # (N, 5, 5, 24)
l_conv9_2 = l_conv9_2.view(batch_size, -1, 4) # (N, 150, 4)
l_conv10_2 = self.loc_conv10_2(conv10_2_feats) # (N, 16, 3, 3)
l_conv10_2 = l_conv10_2.permute(0, 2, 3, 1).contiguous() # (N, 3, 3, 16)
l_conv10_2 = l_conv10_2.view(batch_size, -1, 4) # (N, 36, 4)
l_conv11_2 = self.loc_conv11_2(conv11_2_feats) # (N, 16, 1, 1)
l_conv11_2 = l_conv11_2.permute(0, 2, 3, 1).contiguous() # (N, 1, 1, 16)
l_conv11_2 = l_conv11_2.view(batch_size, -1, 4) # (N, 4, 4)
# Predict classes in localization boxes
c_conv4_3 = self.cl_conv4_3(conv4_3_feats) # (N, 4 * n_classes, 38, 38)
c_conv4_3 = c_conv4_3.permute(0, 2, 3,
1).contiguous() # (N, 38, 38, 4 * n_classes), to match prior-box order (after .view())
c_conv4_3 = c_conv4_3.view(batch_size, -1, self.n_classes) # (N, 5776, n_classes), there are a total 5776 boxes on this feature map
c_conv7 = self.cl_conv7(conv7_feats) # (N, 6 * n_classes, 19, 19)
c_conv7 = c_conv7.permute(0, 2, 3, 1).contiguous() # (N, 19, 19, 6 * n_classes)
c_conv7 = c_conv7.view(batch_size, -1, self.n_classes) # (N, 2166, n_classes), there are a total 2166 boxes on this feature map
c_conv8_2 = self.cl_conv8_2(conv8_2_feats) # (N, 6 * n_classes, 10, 10)
c_conv8_2 = c_conv8_2.permute(0, 2, 3, 1).contiguous() # (N, 10, 10, 6 * n_classes)
c_conv8_2 = c_conv8_2.view(batch_size, -1, self.n_classes) # (N, 600, n_classes)
c_conv9_2 = self.cl_conv9_2(conv9_2_feats) # (N, 6 * n_classes, 5, 5)
c_conv9_2 = c_conv9_2.permute(0, 2, 3, 1).contiguous() # (N, 5, 5, 6 * n_classes)
c_conv9_2 = c_conv9_2.view(batch_size, -1, self.n_classes) # (N, 150, n_classes)
c_conv10_2 = self.cl_conv10_2(conv10_2_feats) # (N, 4 * n_classes, 3, 3)
c_conv10_2 = c_conv10_2.permute(0, 2, 3, 1).contiguous() # (N, 3, 3, 4 * n_classes)
c_conv10_2 = c_conv10_2.view(batch_size, -1, self.n_classes) # (N, 36, n_classes)
c_conv11_2 = self.cl_conv11_2(conv11_2_feats) # (N, 4 * n_classes, 1, 1)
c_conv11_2 = c_conv11_2.permute(0, 2, 3, 1).contiguous() # (N, 1, 1, 4 * n_classes)
c_conv11_2 = c_conv11_2.view(batch_size, -1, self.n_classes) # (N, 4, n_classes)
# A total of 8732 boxes (8732 = 5776 + 2166 + 600 + 150 + 36 + 4)
# Concatenate in this specific order (i.e. must match the order of the prior-boxes)
locs = torch.cat([l_conv4_3, l_conv7, l_conv8_2, l_conv9_2, l_conv10_2, l_conv11_2], dim=1) # (N, 8732, 4)
classes_scores = torch.cat([c_conv4_3, c_conv7, c_conv8_2, c_conv8_2, c_conv10_2, c_conv11_2], dim=1) # (N, 8732, n_classes)
return locs, classes_scores
class SSD300(nn.Module):
"""
The SSD300 Network - encapsulates the base VGG network, auxiliary, and prediction convolutions.
"""
def __init__(self, n_classes):
super(SSD300, self).__init__()
self.n_classes = n_classes
self.base = VGGBase()
self.aux_convs = AuxiliaryConvolutions()
self.pred_convs = PredictionConvolutions(n_classes)
# Since lower level features (conv4_3 feats) have considerably larger scales, we take the L2 norm and rescale
# Rescale factor is initially set at 20, but is learned for each channel during back-prop
self.rescale_factors = nn.Parameter(torch.FloatTensor(1, 512, 1, 1)) # there are 512 channels in conv4_3_feats
nn.init.constant_(self.rescale_factors, 20)
# Prior boxes
self.priors_cxcy = self.create_prior_boxes()
def forward(self, image):
"""
Forward propagation.
:param image: images, a tensor of dimensions (N, 3, 300, 300)
:return: 8732 locations and class scores (i.e. w.r.t each prior box) for each image
"""
# Run VGG base network convolutions (lower level feature map generators)
conv4_3_feats, conv7_feats = self.base(image) # (N, 512, 38, 38), (N, 1024, 19, 19)
# Rescale conv4_3 after L2 norm
norm = conv4_3_feats.pow(2).sum(dim=1, keepdim=True).sqrt() # (N, 1, 38, 38)
conv4_3_feats = conv4_3_feats / norm # (N, 512, 38, 38)
conv4_3_feats = conv4_3_feats * self.rescale_factors # (N, 512, 38, 38)
# (PyTorch autobroadcasts singleton dimensions during arithmetic)
# Run auxiliary convolutions (higher level feature map generators)
conv8_2_feats, conv9_2_feats, conv10_2_feats, conv11_2_feats = self.aux_convs(conv7_feats)
# (N, 512, 10, 10), (N, 256, 5, 5), (N, 256, 3, 3), (N, 256, 1, 1)
# Run prediction convolutions (predict offsets w.r.t prior-boxes and classes in each resulting localization box)
locs, classes_scores = self.pred_convs(conv4_3_feats, conv7_feats, conv8_2_feats, conv9_2_feats, conv10_2_feats, conv11_2_feats)
# (N, 8732, 4), (N, 8732, n_classes)
return locs, classes_scores
def create_prior_boxes(self):
"""
Create the 8732 prior (default) boxes for the SSD300
:return: prior boxes in center-size coordinates, a tensor of dimensions (8732, 4)
"""
fmap_dims = {'conv4_3': 38,
'conv7': 19,
'conv8_2': 10,
'conv9_2': 5,
'conv10_2': 3,
'conv11_2': 1}
obj_scales = {'conv4_3': 0.1,
'conv7': 0.2,
'conv8_2': 0.375,
'conv9_2': 0.55,
'conv10_2': 0.725,
'conv11_2': 0.9}
aspect_ratios = {'conv4_3': [1., 2., 0.5],
'conv7': [1., 2., 3., 0.5, .333],
'conv8_2': [1., 2., 3., 0.5, .333],
'conv9_2': [1., 2., 3., 0.5, .333],
'conv10_2': [1., 2., 0.5],
'conv11_2': [1., 2., 0.5]}
fmaps = list(fmap_dims.keys())
prior_boxes = []
for k, fmap in enumerate(fmaps):
for i in range(fmap_dims[fmap]):
for j in range(fmap_dims[fmap]):
cx = (j + 0.5) / fmap_dims[fmap]
cy = (i + 0.5) / fmap_dims[fmap]
for ratio in aspect_ratios[fmap]:
prior_boxes.append([cx, cy, obj_scales[fmap] * sqrt(ratio), obj_scales[fmap] / sqrt(ratio)]) # (cx, cy, w, h)
# For an aspect ratio of 1, use an additional prior whose scale is the geometric mean of the
# scale of the current feature map and the scale of the next feature map
if ratio == 1.:
try:
additional_scale = sqrt(obj_scales[fmap] * obj_scales[fmaps[k + 1]])
# For the last feature map, there is no "next" feature map
except IndexError:
additional_scale = 1.
prior_boxes.append([cx, cy, additional_scale, additional_scale])
prior_boxes = torch.FloatTensor(prior_boxes).to(device) # (8732, 4)
prior_boxes.clamp_(0, 1) # (8732, 4)
return prior_boxes
def detect_objects(self, predicted_locs, predicted_scores, min_score, max_overlap, top_k):
"""
Decipher the 8732 locations and class scores (output of this SSD300) to detect objects.
For each class, perform Non-Maximum Suppression (NMS) on boxes that are above a minimum threshold.
:param predicted_locs: predicted locations/boxes w.r.t the 8732 prior boxes, a tensor of dimensions (N, 8732, 4)
:param predicted_scores: class scores for each of the encoded locations/boxes, a tensor of dimensions (N, 8732, n_classes)
:param min_score: minimum threshold for a box to be considered a match for a certain class
:param max_overlap: maximum overlap two boxes can have so that the one with the lower score is not suppressed via NMS
:param top_k: if there are a lot of resulting detection across all classes, keep only the top 'k'
:return: detections (boxes, labels, and scores), lists of length batch_size
"""
batch_size = predicted_locs.size(0)
n_priors = self.priors_cxcy.size(0)
predicted_scores = F.softmax(predicted_scores, dim=2) # (N, 8732, n_classes)
# Lists to store final predicted boxes, labels, and scores for all images
all_images_boxes = list()
all_images_labels = list()
all_images_scores = list()
assert n_priors == predicted_locs.size(1) == predicted_scores.size(1)
for i in range(batch_size):
# Decode object coordinates from the form we regressed predicted boxes to
decoded_locs = cxcy_to_xy(
gcxgcy_to_cxcy(predicted_locs[i], self.priors_cxcy)) # (8732, 4), these are fractional pt. coordinates
# Lists to store boxes and scores for this image
image_boxes = list()
image_labels = list()
image_scores = list()
max_score, best_label = predicted_scores[i].max(dim=1) # (8732)
# Check for each class
for c in range(1, self.n_classes):
# Keep only predicted boxes and scores where scores for this class are above minimum score
class_scores = predicted_scores[i][:, c] # (8732)
score_above_min_score = class_scores > min_score # torch.uint8 (byte) tensor, for indexing
n_above_min_score = score_above_min_score.sum().item()
if n_above_min_score == 0:
continue
class_scores = class_scores[score_above_min_score] # (n_qualified), n_min_score <= 8732
class_decoded_locs = decoded_locs[score_above_min_score] # (n_qualified, 4)
# Sort predicted boxes and scores by scores
class_scores, sort_ind = class_scores.sort(dim=0, descending= True) # (n_qualified), (n_min_score)
class_decoded_locs = class_decoded_locs[sort_ind] # (n_min_score, 4)
# Find the overlap between predicted boxes
overlap = find_jaccard_overlap(class_decoded_locs, class_decoded_locs) # (n_qualified, n_min_score)
# Non-Maximum Suppression (NMS)
# A torch.unit8 (byte) tensor to keep track of which predicted boxes to suppress
# 1 implies suppress, 0 implies don't suppress
suppress = torch.zeros((n_above_min_score), dtype=torch.uint8).to(device) # (n_qulified)
# Consider each box in order of decreasing scores
for box in range(class_decoded_locs.size(0)):
# If this box is aleready marked for suppression
if suppress[box] == 1:
continue
# Suppress boxes whose overlaps (with this box) are greater than maximum overlap
# Find such boxes and update suppress indices
suppress = torch.max(suppress, overlap[box] > max_overlap)
# The max operation retains previously suppressed boxes, like an 'OR' operation
# Don't suppress this box, even though it has and overlap of 1 with itself
suppress[box] = 0
# Store only unsuppressed boxes for this class
image_boxes.append(class_decoded_locs[1 - suppress])
image_labels.append(torch.LongTensor((1 - suppress).sum().item() * [c]).to(device))
image_scores.append(class_scores[1 - suppress])
# If no object in any class is found, store a placeholder for 'background'
if len(image_boxes) == 0:
image_boxes.append(torch.FloatTensor([0., 0., 1., 1.]).to(device))
image_labels.append(torch.LongTensor([0]).to(device))
image_scores.append(torch.FloatTensor([0.]).to(device))
# Concatenate into single tensors
image_boxes = torch.cat(image_boxes, dim=0) # (n_objects, 4)
image_labels = torch.cat(image_labels, dim=0) # (n_objects)
image_scores = torch.cat(image_scores, dim=0) # (n_objects)
n_objects = image_scores.size(0)
# keep only the top k objects
if n_objects > top_k:
image_scores, sort_ind = image_scores.sort(dim=0, descending= True)
image_scores = image_scores[:top_k] # (top_k)
image_boxes = image_boxes[sort_ind][:top_k] # (top_k, 4)
image_labels = image_labels[sort_ind][:top_k] # (top_k)
# Append to lists that store predicted boxes and scores for all images
all_images_boxes.append(image_boxes)
all_images_labels.append(image_labels)
all_images_scores.append(image_scores)
return all_images_boxes, all_images_labels, all_images_scores # lists of length batch_size
Raw
ssd detector_utils.py
import json
import os
import torch
import random
import xml.etree.ElementTree as ET
import torchvision.transforms.functional as FT
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
## Label map
voc_labels = ('aeroplane', 'bicycle', 'bird', 'boat', 'bottle', 'bus', 'car', 'cat', 'chair', 'cow', 'diningtable',
'dog', 'horse', 'motorbike', 'person', 'pottedplant', 'sheep', 'sofa', 'train', 'tvmonitor')
print('Length of labels : {}'.format(len(voc_labels)))
label_map = {k: v+1 for v, k in enumerate(voc_labels)}
label_map['background'] = 0
rev_label_map = {v: k for k, v in label_map.items()} # Inverse mapping
print(rev_label_map)
# Color map for bounding boxes of detected objects
# from https://sashat.me/2017/01/11/list-of-20-simple-distinct-colors/
distinct_colors = ['#e6194b', '#3cb44b', '#ffe119', '#0082c8', '#f58231', '#911eb4', '#46f0f0', '#f032e6',
'#d2f53c', '#fabebe', '#008080', '#000080', '#aa6e28', '#fffac8', '#800000', '#aaffc3', '#808000',
'#ffd8b1', '#e6beff', '#808080', '#FFFFFF']
label_color_map = {k: distinct_colors[i] for i, k in enumerate(label_map.keys())}
print(label_color_map)
##
def parse_annotation(annotation_path):
tree = ET.parse(annotation_path)
root = tree.getroot()
boxes = list()
labels = list()
difficulties = list()
for object in root.iter('object'):
difficult = int(object.find('difficult').text == '1')
label = object.find('name').text.lower().strip()
if label not in label_map:
continue
bbox = object.find('bndbox')
xmin = int(bbox.find('xmin').text) - 1
ymin = int(bbox.find('ymin').text) - 1
xmax = int(bbox.find('xmax').text) - 1
ymax = int(bbox.find('ymax').text) - 1
boxes.append([xmin, ymin, xmax, ymax])
labels.append(label_map[label])
difficulties.append(difficult)
return {'boxes' : boxes,
'labels' : labels,
'difficulties' : difficulties}
def create_data_lists(voc07_path, voc12_path, output_folder):
"""
Create lists of images, the bounding boxes and labels of the objects in these images, and save these to file.
:param voc07_path: path to the 'VOC2007' Folder
:param voc12_path: path to the 'VOC2012' Folder
:param output_folder: folder where the JSONs must be saved
"""
voc07_path = os.path.abspath(voc07_path)
voc12_path = os.path.abspath(voc12_path)
train_images = list()
train_objects = list()
n_objects = 0
# Traning data
for path in [voc07_path, voc12_path]:
# Find IDs of images in traning data
with open(os.path.join(path, 'ImageSets/Main/trainval.txt')) as f:
ids = f.read().splitlines()
for id in ids:
# Parse annotation's XML file
objects = parse_annotation(os.path.join(path, 'Annotations', id + '.xml'))
if len(objects) == 0:
continue
n_objects += len(objects)
train_objects.append(objects)
train_images.append(os.path.join(path, 'JPEGImages', id + '.jpg'))
assert len(train_objects) == len(train_images)
# Save to file
with open(os.path.join(output_folder, 'TRAIN_images.json'), 'w') as j:
json.dump(train_images, j)
with open(os.path.join(output_folder, 'TRAIN_objects.json'), 'w') as j:
json.dump(train_objects, j)
with open(os.path.join(output_folder, 'label_map.json'), 'w') as j:
json.dump(label_map, j)
print('\nThere are %d training images containing a total of %d objects. Files have been saved to %s.' % (
len(train_images), n_objects, os.path.abspath(output_folder)))
# Validation data
test_images = list()
test_objects = list()
n_objects = 0
# Find IDs of images in validation data
with open(os.path.join(voc07_path, 'ImageSets/Main/test.txt')) as f:
ids = f.read.splitlines()
for id in ids:
# Parse annotation's XML files
objects = parse_annotation(os.path.join(voc07_path, 'Annotations', id + '.xml'))
if len(objects) == 0:
continue
test_objects.append(objects)
n_objects += len(objects)
test_images.append(os.path.join(voc07_path, 'JPEGImages', id + '.jpg'))
assert len(test_images) == len(test_objects)
# Save to file
with open(os.path.join(output_folder, 'TEST_images.json'), 'w') as j:
json.dump(test_images, j)
with open(os.path.join(output_folder, 'TEST_objects.json'), 'w') as j:
json.dump(test_objects, j)
print('\nThere are %d validation images containing a total of %d objects. Files have been saved to %s.' % (
len(test_images), n_objects, os.path.abspath(output_folder)))
def decimate(tensor, m):
"""
Decimate a tensor by a factor 'm', i.e. downsample by keeping every 'm'th value.
This is used when we convert FC layers to equivalent Convolutional layers, BUT of a smaller size.
:param tensor: tensor to be decimated
:param m: list of decimation factors for each dimension of the tensor; None if Not to be decimated along a dimension
:return: decimated tensor
"""
assert tensor.dim() == len(m)
for d in range(tensor.dim()):
if m[d] is not None:
tensor = tensor.index_select(dim=d, index=torch.arange(start=0, end=tensor.size(d), step= m[d]).long())
return tensor
def photometric_distort(image):
"""
Distort brightness, contrast, saturation, and hue, each with a 50% chance, in random order.
:param image: image, a PIL Image
:return: distorted image
"""
new_image = image
distortions = [FT.adjust_brightness,
FT.adjust_contrast,
FT.adjust_saturation,
FT.adjust_hue]
random.shuffle(distortions)
for d in distortions:
if random.random() < 0.5:
if d.__name__ is 'adjust_hue':
# Caffe repo uses a 'hue_delta' of 18 - we divide by 255 because PyTorch needs a normalized value
adjust_factor = random.uniform(-18 / 255., 18 / 255.)
else:
# Caffe repo uses 'lower' and 'upper' values of 0.5 and 1.5 for brightness, contrast, and saturation
adjust_factor = random.uniform(0.5, 1.5)
# Apply this distortion
new_image = d(new_image, adjust_factor)
return new_image
def cxcy_to_xy(cxcy):
"""
Convert bounding boxes from center-size coordinates (c_x, c_y, w, h) to boundary coordinates (x_min, y_min, x_max, y_max)
:param cxcy: bounding boxes in center-size coordinates, a tensor of size (n_boxes, 4)
:return: bounding boxes in boundary coordinates, a tensor of size (n_boxes, 4)
"""
return torch.cat([cxcy[:, :2] - (cxcy[:, 2:] / 2), # x_min, y_min
cxcy[:, :2] + (cxcy[:, 2:] / 2)], 1) # x_max, y_max
def gcxgcy_to_cxcy(gcxgcy, priors_cxcy):
"""
Decode bounding box coordinates predicted by the model, since they are encoded in the form mentioned above.
They are decoded into center-size coordinates.
This is the inverse of the function above.
:param gcxgcy: encoded bounding boxes, i.e output of the model, a tensor of size (n_priors, 4)
:param priors_cxcy: prior boxes with respect to which the encoding is defined, a tensor of size (n_priors, 4)
:return: decoded bounding boxes in center-size form, a tensor of size (n_priors, 4)
"""
return torch.cat([gcxgcy[:, :2] * priors_cxcy[:, 2:] / 10 + priors_cxcy[:, :2], # c_x, c_y
torch.exp(gcxgcy[:, 2:] / 5) * priors_cxcy[:, 2:]], 1) # w, h
def expand(image, boxes, filler):
"""
Perform a zooming out operation by placing the image in a larger canvas of filler material.
:param image: image, a tensor of dimensions (3, original_h, original_w)
:param boxes: bounding boxes in boundary coordinates, a tensor of dimensions (n_objects, 4)
:param filler: RBG values of the filler material, a list like [R, G, B]
:return: expanded image, updated bounding box coordinates
"""
# Calculate dimensions of proposed expanded (zoomed-out) image
original_h = image.size(1)
original_w = image.size(2)
max_scale = 4
scale = random.uniform(1, max_scale)
new_h = int(scale * original_h)
new_w = int(scale * original_w)
# Create such an image with the filler
filler = torch.FloatTensor(filler) # (3)
new_image = torch.ones((3, new_h, new_w), dtype=torch.float) * filler.unsqueeze(1).unsqueeze(1) # (3, new_h, new_w)
# Note - do not use expand() list new_image = filler.unsqueeze(1).unsqueeze(1).expand(3, new_h, new_w)
# because all expanded values will share the same memory, so changing on pixel will change all
# Place the original image at random coordinates in this new image (origin at top-left of image)
left = random.randint(0, new_w - original_w)
right = left + original_w
top = random.randint(0, new_h - original_h)
bottom = top + original_h
new_image[:, top:bottom, left::right] = image
# Adjust bounding boxes coordinates accordingly
new_boxes = boxes + torch.FloatTensor([left, top, left, top]).unsqueeze(0) # (n_objects, 4)
return new_image, new_boxes
def find_intersection(set_1, set_2):
"""
Find the intersection of every box combination between two sets of boxes that are in boundary coordinates.
:param set_1: set 1, a tensor of dimensions (n1, 4)
:param set2: set 2, a tensor of dimensions (n2, 4)
:return: intersection of each of the boxes in set 1 with respect to each of the boxes in set 2, a tensor of dimension (n1, n2)
"""
# PyTorch auto-broadcast singleton dimensions
lower_bounds = torch.max(set_1[:, :2].unsqueeze(1), set_2[:, :2].unsqueeze(0)) # (n1, n2, 2) 2 : inter_min_x, inter_min_y
upper_bounds = torch.min(set_1[:, 2:].unsqueeze(1), set_2[:, 2:].unsqueeze(0)) # (n1, n2, 2) 2 : inter_max_x, inter_max_y
intersection_dims = torch.clamp(upper_bounds - lower_bounds, min=0) # (n1, n2, 2)
return intersection_dims[:, :, 0] * intersection_dims[:, :, 1] # (n1, n2) # Calculate Area
def find_jaccard_overlap(set_1, set_2):
"""
Find the Jaccard Overlap (IoU) of every box combination between two sets of boxes that are in boundary coordinates.
:param set_1: set 1, a tensor of dimensions (n1, 4)
:param set_2: set 2, a tensor of dimensions (n2, 4)
:return: Jaccard Overlap of each of the boxes in set 1 with respect to each of the boxes in set 2, a tensor of dimensions (n1, n2)
"""
# Find intersections
intersection = find_intersection(set_1, set_2) # (n1, n2)
# Find areas of each box in both sets
areas_set_1 = (set_1[:, 2] - set_1[:, 0]) * (set_1[:, 3] - set_1[:, 1]) # (n1)
areas_set_2 = (set_2[:, 2] - set_2[:, 0]) * (set_2[:, 3] - set_2[:, 2]) # (n2)
# Find the union
# PyTorch auto-broadcasts singleton dimensions
union = areas_set_1.unsqueeze(1) + areas_set_2.unsqueeze(0) - intersection # (n1,n2)
iou = intersection / union # (n1, n2)
return iou
def random_crop(image, boxes, labels, difficulties):
"""
Performs a random crop in the manner stated in the paper. Helps to learn to detect larger and partial objects.
Note that some objects may be cut out entirely
:param image: image, a tensor of dimensions (3, original_h, original_w)
:param boxes: bounding boxes in boundary coordinates, a tensor of dimensions (n_objects, 4)
:param labels: labels of objects, a tensor of dimensions (n_objects)
:param difficulties: difficulties of detection of these objects, a tensor of dimensions (n_objects)
:return: cropped image, updated bounding box coordinates, updated labels, updated difficulties
"""
original_h = image.size(1)
original_w = image.size(2)
# Keep choosing aa minimum overlap until a successful crop is made
while True:
# Randomly draw the value for minimum overlap
min_overlap = random.choice([0., .1, .3, .5, .7, .9, None]) # 'None' refer to no cropping
# If not cropping
if min_overlap is None:
return image, boxes, labels, difficulties
# Try up to 50 times for this choice of minimum overlap
# This isn't mentioned in the paper, but 50 is chosen in paper author's Caffe repo
max_trials = 50
for _ in range(max_trials):
# Crop dimensions must be in [0.3, 1] of original dimensions
# Note - it's [0.1, 1] in the paper, but actually [0.3, 1] in the author's repo
min_scale = 0.3
scale_h = random.uniform(min_scale, 1)
scale_w = random.uniform(min_scale, 1)
new_h = int(scale_h * original_h)
new_w = int(scale_w * original_w)
# Aspect ratio has to be in [0.5, 2]
aspect_ratio = new_h / new_w
if not 0.5 < aspect_ratio < 2:
continue
# Crop coordinates (origin at top-left of image)
left = random.randint(0, original_w - new_w)
right = left + new_w
top = random.randint(0, original_h - new_h)
bottom = top + new_h
crop = torch.FloatTensor([left, top, right, bottom]) # (4)
# Calculate Jaccard overlap between the crop and the bounding boxes
overlap = find_jaccard_overlap(crop.unsqueeze(0), boxes) # (1, n_objects), n_objects is the no. of objects in this image
overlap = overlap.squeeze(0) # (n_objects)
# If not a single bounding box has a Jaccard overlap of greater than the minimum, try again
if overlap.max().item() < min_overlap:
continue
# Crop image
new_image = image[:, top:bottom, left:right] # (3, new_h, new_w)
# Find centers of original bounding boxes
bb_centers = (boxes[:, :2] + boxes[:, 2:]) / 2. # (n_objects, 2)
# Find bounding boxes whose centers are in the crop
centers_in_crop = (bb_centers[:, 0] > left) * (bb_centers[:, 0] < right) * (bb_centers[:, 1] > top) * (bb_centers[:, 1] < bottom) # (n_objects)
# If not a single bounding box has its center in the crop, try again
if not centers_in_crop.any():
continue
# Discard bounding boxes that don't meet this criterion
new_boxes = boxes[centers_in_crop, :1]
new_labels = labels[centers_in_crop]
new_difficulties = difficulties[centers_in_crop]
# Calculate bounding boxes new coordinates in the crop
new_boxes[:, :2] = torch.max(new_boxes[:, :2], crop[:2]) # crop[:2] is [left, top]
new_boxes[:, :2] -= crop[:2]
new_boxes[:, 2:] = torch.min(new_boxes[:, 2:], crop[2:]) # crop[2:] is [right, bottom]
new_boxes[:, 2:] -= crop[:2]
return new_image, new_boxes, new_labels, new_difficulties
def flip(image, boxes):
"""
Flip image horizontally.
:param image: image, a PIL image
:param boxes: bounding boxes in boundary coordinates, a tensor of dimensions (n_objects, 4)
:return: flipped image, updated bounding box coordinates
"""
# Flip image
new_image = FT.hflip(image)
# Flip boxes
new_boxes = boxes
new_boxes[:, 0] = image.width - boxes[:, 0] - 1
new_boxes[:, 2] = image.width - boxes[:, 2] - 1
new_boxes = new_boxes[:, [2, 1, 0, 3]]
return new_image, new_boxes
def resize(image, boxes, dims=(300,300), return_percent_coords=True):
"""
Resize image. For the SSD300, resize to (300, 300)
Since percent/fractional coordinates are calculated for the bounding boxes (w.r.t image dimensions) in this process,
you may choose to retain them.
:param image: image, a PIL Image
:param boxes: bounding boxes in boundary coordinates, a tensor of dimensions (n_objects, 4)
:return: resized image, updated bounding box coordinates (or fractional coordinates, in which case they remain the same)
"""
# Resize image
new_image = FT.resize(image, dims)
# Resize bounding boxes
old_dims = torch.FloatTensor([image.width, image.height, image.width, image.height]).unsqueeze(0)
new_boxes = boxes / old_dims # percent coordinates
if not return_percent_coords:
new_dims = torch.FloatTensor([dims[1], dims[0], dims[1], dims[0]]).unsqueeze(0)
new_boxes = new_boxes * new_dims
return new_image, new_boxes
def transform(image, boxes, labels, difficulties, split):
"""
Apply the transformations above.
:param image: image, a PIL Image
:param boxes: bounding boxes in boundary coordinates, a tensor of dimensions (n_objects, 4)
:param labels: labels of objects, a tensor of dimensions (n_objects)
:param difficulties: difficulties of detection of these objects, a tensor of dimensions (n_objects)
:param split: one of 'TRAIN' or 'TEST', since different sets of transformations are applied
:return: transformed image, transformed bounding box coordinates, transformed labels, transformed difficulties
"""
assert split in {'TRAIN', 'TEST'}
# Mean and standard deviation of ImageNet data that our base VGG from torchvision was trained on
mean = [0.485, 0.456, 0.406]
std = [0.229, 0.224, 0.225]
new_image = image
new_boxes = boxes
new_labels = labels
new_difficulties = difficulties
# Skip the following operations if validation/evaluation
if split == 'TRAIN':
# A series of photometic distortions in random order, each with 50% chance of occurrence, as in Caffe repo
new_image = photometric_distort(new_image)
# Convert PIL image to Torch tensor
new_image = FT.to_tensor(new_image)
# Expand image (zoom out) with a 50% chance - helpful for training detection of small objects
# Fill surrounding space with the mean of ImageNet data that our base VGG was trained on
if random.random < 0.5:
new_image, new_boxes = expand(new_image, filler=mean)
# Randomly crop image (zoom in)
new_image, new_boxes, new_labels, new_difficulties = random_crop(new_image, new_boxes, new_labels, new_difficulties)
# Convert Torch tensor to PIL image
new_image = FT.to_pil_image(new_image)
# Flip image with a 50% chance
if random.random() < 0.5:
new_image, new_boxes = flip(new_image, new_boxes)
# Resize image to (300, 300) - this also converts absolute boundary coordinates to their fractional form
new_image, new_boxes = resize(new_image, new_boxes, dims=(300,300))
# Convert PIL image to Torch tensor
new_image = FT.to_tensor(new_image)
# Normalize by mean and standard deviation of ImageNet
new_image = FT.normalize(new_image, mean=mean, std=std)
return new_image, new_boxes, new_labels, new_difficulties
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment