File size: 11,773 Bytes
ca188f3 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 |
#!/usr/bin/env python3
"""
Creates a Pytorch dataset to load the Pascal VOC & MS COCO datasets
"""
# Standard Library Imports
import os
import random
# Third-Party Imports
import torch
import numpy as np
import pandas as pd
from PIL import Image, ImageFile
from torch.utils.data import Dataset, DataLoader
# Local Imports
import config
from utils import xywhn2xyxy, xyxy2xywhn
from utils import (
cells_to_bboxes,
iou_width_height as iou,
non_max_suppression as nms,
plot_image
)
ImageFile.LOAD_TRUNCATED_IMAGES = True
class YOLODataset(Dataset):
def __init__(
self,
csv_file,
img_dir,
label_dir,
anchors,
image_size=416,
S=[13, 26, 52],
C=20,
transform=None,
mosaic=0.75
):
self.annotations = pd.read_csv(csv_file)
self.img_dir = img_dir
self.label_dir = label_dir
self.image_size = image_size
self.mosaic_border = [image_size // 2, image_size // 2]
self.transform = transform
self.S = S
self.anchors = torch.tensor(anchors[0] + anchors[1] + anchors[2]) # for all 3 scales
self.num_anchors = self.anchors.shape[0]
self.num_anchors_per_scale = self.num_anchors // 3
self.C = C
self.ignore_iou_thresh = 0.5
self.mosaic = mosaic
def __len__(self):
return len(self.annotations)
def load_images(self, index):
"""
Function to load the images
Args:
index: Image index in the dataset
Returns: Images and their bounding boxes
"""
label_path = os.path.join(self.label_dir, self.annotations.iloc[index, 1])
bboxes = np.roll(np.loadtxt(fname=label_path, delimiter=" ", ndmin=2), 4, axis=1).tolist()
img_path = os.path.join(self.img_dir, self.annotations.iloc[index, 0])
img = np.array(Image.open(img_path).convert("RGB"))
return img, bboxes
def load_mosaic(self, index, p=0.75):
"""
Args:
p: Percentage of images for which mosaic augmentation to be applied
index: Index of the images in the dataset
Returns: Images, Labels
"""
# Call Normal images and labels 25% of the time
if random.random() >= p:
return self.load_images(index)
# YOLOv5 4-mosaic loader. Loads 1 image + 3 random images into a 4-image mosaic
labels4 = []
s = self.image_size
yc, xc = (int(random.uniform(x, 2 * s - x)) for x in self.mosaic_border) # mosaic center x, y
indices = [index] + random.choices(range(len(self)), k=3) # 3 additional image indices
random.shuffle(indices)
for i, index in enumerate(indices):
# Load image
label_path = os.path.join(self.label_dir, self.annotations.iloc[index, 1])
bboxes = np.roll(np.loadtxt(fname=label_path, delimiter=" ", ndmin=2), 4, axis=1).tolist()
img_path = os.path.join(self.img_dir, self.annotations.iloc[index, 0])
img = np.array(Image.open(img_path).convert("RGB"))
h, w = img.shape[0], img.shape[1]
labels = np.array(bboxes)
# place img in img4
if i == 0: # top left
img4 = np.full((s * 2, s * 2, img.shape[2]), 114, dtype=np.uint8) # base image with 4 tiles
x1a, y1a, x2a, y2a = max(xc - w, 0), max(yc - h, 0), xc, yc # xmin, ymin, xmax, ymax (large image)
x1b, y1b, x2b, y2b = w - (x2a - x1a), h - (y2a - y1a), w, h # xmin, ymin, xmax, ymax (small image)
elif i == 1: # top right
x1a, y1a, x2a, y2a = xc, max(yc - h, 0), min(xc + w, s * 2), yc
x1b, y1b, x2b, y2b = 0, h - (y2a - y1a), min(w, x2a - x1a), h
elif i == 2: # bottom left
x1a, y1a, x2a, y2a = max(xc - w, 0), yc, xc, min(s * 2, yc + h)
x1b, y1b, x2b, y2b = w - (x2a - x1a), 0, w, min(y2a - y1a, h)
elif i == 3: # bottom right
x1a, y1a, x2a, y2a = xc, yc, min(xc + w, s * 2), min(s * 2, yc + h)
x1b, y1b, x2b, y2b = 0, 0, min(w, x2a - x1a), min(y2a - y1a, h)
img4[y1a:y2a, x1a:x2a] = img[y1b:y2b, x1b:x2b] # img4[ymin:ymax, xmin:xmax]
padw = x1a - x1b
padh = y1a - y1b
# Labels
if labels.size:
labels[:, :-1] = xywhn2xyxy(labels[:, :-1], w, h, padw, padh) # normalized xywh to pixel xyxy format
labels4.append(labels)
# Concat/clip labels
labels4 = np.concatenate(labels4, 0)
for x in (labels4[:, :-1],):
np.clip(x, 0, 2 * s, out=x) # clip when using random_perspective()
# img4, labels4 = replicate(img4, labels4) # replicate
labels4[:, :-1] = xyxy2xywhn(labels4[:, :-1], 2 * s, 2 * s)
labels4[:, :-1] = np.clip(labels4[:, :-1], 0, 1)
labels4 = labels4[labels4[:, 2] > 0]
labels4 = labels4[labels4[:, 3] > 0]
return img4, labels4
def __getitem__(self, index):
image, bboxes = self.load_mosaic(index, p=self.mosaic)
if self.transform:
augmentations = self.transform(image=image, bboxes=bboxes)
image = augmentations["image"]
bboxes = augmentations["bboxes"]
# Below assumes 3 scale predictions (as paper) and same num of anchors per scale
targets = [torch.zeros((self.num_anchors // 3, S, S, 6)) for S in self.S]
for box in bboxes:
iou_anchors = iou(torch.tensor(box[2:4]), self.anchors)
anchor_indices = iou_anchors.argsort(descending=True, dim=0)
x, y, width, height, class_label = box
has_anchor = [False] * 3 # each scale should have one anchor
for anchor_idx in anchor_indices:
scale_idx = anchor_idx // self.num_anchors_per_scale
anchor_on_scale = anchor_idx % self.num_anchors_per_scale
S = self.S[scale_idx]
i, j = int(S * y), int(S * x) # which cell
anchor_taken = targets[scale_idx][anchor_on_scale, i, j, 0]
if not anchor_taken and not has_anchor[scale_idx]:
targets[scale_idx][anchor_on_scale, i, j, 0] = 1
x_cell, y_cell = S * x - j, S * y - i # both between [0,1]
width_cell, height_cell = (
width * S,
height * S,
) # can be greater than 1 since it's relative to cell
box_coordinates = torch.tensor(
[x_cell, y_cell, width_cell, height_cell]
)
targets[scale_idx][anchor_on_scale, i, j, 1:5] = box_coordinates
targets[scale_idx][anchor_on_scale, i, j, 5] = int(class_label)
has_anchor[scale_idx] = True
elif not anchor_taken and iou_anchors[anchor_idx] > self.ignore_iou_thresh:
targets[scale_idx][anchor_on_scale, i, j, 0] = -1 # ignore prediction
return image, tuple(targets)
# class Mosaic(object):
# """
# Class for custom augmentation strategy
# """
# def __init__(self, image_size):
# """
# Constructor
# """
# self.image_size = image_size
# self.mosaic_border = [image_size // 2, image_size // 2]
# self.annotations = pd.read_csv(csv_file)
# self.img_dir = img_dir
# self.label_dir = label_dir
#
# def __len__(self):
# return len(self.annotations)
#
# def __call__(self, index):
# """
# Augmentation strategy to be implemented when called
# """
# # YOLOv5 4-mosaic loader. Loads 1 image + 3 random images into a 4-image mosaic
# labels4 = []
# s = self.image_size
# yc, xc = (int(random.uniform(x, 2 * s - x)) for x in self.mosaic_border) # mosaic center x, y
# indices = [index] + random.choices(range(len(self)), k=3) # 3 additional image indices
# random.shuffle(indices)
# for i, index in enumerate(indices):
# # Load image
# label_path = os.path.join(self.label_dir, self.annotations.iloc[index, 1])
# bboxes = np.roll(np.loadtxt(fname=label_path, delimiter=" ", ndmin=2), 4, axis=1).tolist()
# img_path = os.path.join(self.img_dir, self.annotations.iloc[index, 0])
# img = np.array(Image.open(img_path).convert("RGB"))
#
# h, w = img.shape[0], img.shape[1]
# labels = np.array(bboxes)
#
# # place img in img4
# if i == 0: # top left
# img4 = np.full((s * 2, s * 2, img.shape[2]), 114, dtype=np.uint8) # base image with 4 tiles
# x1a, y1a, x2a, y2a = max(xc - w, 0), max(yc - h, 0), xc, yc # xmin, ymin, xmax, ymax (large image)
# x1b, y1b, x2b, y2b = w - (x2a - x1a), h - (y2a - y1a), w, h # xmin, ymin, xmax, ymax (small image)
# elif i == 1: # top right
# x1a, y1a, x2a, y2a = xc, max(yc - h, 0), min(xc + w, s * 2), yc
# x1b, y1b, x2b, y2b = 0, h - (y2a - y1a), min(w, x2a - x1a), h
# elif i == 2: # bottom left
# x1a, y1a, x2a, y2a = max(xc - w, 0), yc, xc, min(s * 2, yc + h)
# x1b, y1b, x2b, y2b = w - (x2a - x1a), 0, w, min(y2a - y1a, h)
# elif i == 3: # bottom right
# x1a, y1a, x2a, y2a = xc, yc, min(xc + w, s * 2), min(s * 2, yc + h)
# x1b, y1b, x2b, y2b = 0, 0, min(w, x2a - x1a), min(y2a - y1a, h)
#
# img4[y1a:y2a, x1a:x2a] = img[y1b:y2b, x1b:x2b] # img4[ymin:ymax, xmin:xmax]
# padw = x1a - x1b
# padh = y1a - y1b
#
# # Labels
# if labels.size:
# labels[:, :-1] = xywhn2xyxy(labels[:, :-1], w, h, padw, padh) # normalized xywh to pixel xyxy format
# labels4.append(labels)
#
# # Concat/clip labels
# labels4 = np.concatenate(labels4, 0)
# for x in (labels4[:, :-1],):
# np.clip(x, 0, 2 * s, out=x) # clip when using random_perspective()
# # img4, labels4 = replicate(img4, labels4) # replicate
# labels4[:, :-1] = xyxy2xywhn(labels4[:, :-1], 2 * s, 2 * s)
# labels4[:, :-1] = np.clip(labels4[:, :-1], 0, 1)
# labels4 = labels4[labels4[:, 2] > 0]
# labels4 = labels4[labels4[:, 3] > 0]
# return img4, labels4
#
# def __repr__(self):
# """
# Method to print more infor about the strategy
# """
# return f"{self.__class__.__name__}(mean={self.mean}, std={self.std})"
def test():
anchors = config.ANCHORS
transform = config.test_transforms
dataset = YOLODataset(
"COCO/train.csv",
"COCO/images/images/",
"COCO/labels/labels_new/",
S=[13, 26, 52],
anchors=anchors,
transform=transform,
)
S = [13, 26, 52]
scaled_anchors = torch.tensor(anchors) / (
1 / torch.tensor(S).unsqueeze(1).unsqueeze(1).repeat(1, 3, 2)
)
loader = DataLoader(dataset=dataset, batch_size=1, shuffle=True)
for x, y in loader:
boxes = []
for i in range(y[0].shape[1]):
anchor = scaled_anchors[i]
print(anchor.shape)
print(y[i].shape)
boxes += cells_to_bboxes(
y[i], is_preds=False, S=y[i].shape[2], anchors=anchor
)[0]
boxes = nms(boxes, iou_threshold=1, threshold=0.7, box_format="midpoint")
print(boxes)
plot_image(x[0].permute(1, 2, 0).to("cpu"), boxes)
if __name__ == "__main__":
test()
|