Download images and annotations: BCCD dataset.
Split dataset into train, validation and test set
library(tidyverse)
library(platypus)
library(abind)
library(here)
BCCD_path <- here("development/BCCD/")
annot_path <- file.path(BCCD_path, "Annotations/")
images_path <- file.path(BCCD_path, "JPEGImages/")
c("train", "valid", "test") %>%
walk(~ {
dir.create(file.path(BCCD_path, .))
dir.create(file.path(BCCD_path, ., "Annotations/"))
dir.create(file.path(BCCD_path, ., "JPEGImages/"))
})
annot_paths <- list.files(annot_path, full.names = TRUE)
images_paths <- list.files(images_path, full.names = TRUE)
n_samples <- length(annot_paths)
set.seed(111)
train_ids <- sample(1:n_samples, round(0.8 * n_samples))
valid_ids <- sample(setdiff(1:n_samples, train_ids), round(0.19 * n_samples))
test_ids <- setdiff(1:n_samples, c(train_ids, valid_ids))
walk2(c("train", "valid", "test"), list(train_ids, valid_ids, test_ids), ~ {
annots <- annot_paths[.y]
images <- images_paths[.y]
dir_name <- .x
annots %>% walk(~ file.copy(., gsub("BCCD", paste0("BCCD/", dir_name), .)))
images %>% walk(~ file.copy(., gsub("BCCD", paste0("BCCD/", dir_name), .)))
})Generate custom anchor boxes:
blood_labels <- c("Platelets", "RBC", "WBC")
n_class <- length(blood_labels)
net_h <- 416 # Must be divisible by 32
net_w <- 416 # Must be divisible by 32
anchors_per_grid <- 3
blood_anchors <- generate_anchors(
anchors_per_grid = anchors_per_grid, # Number of anchors (per one grid) to generate
annot_path = annot_path, # Annotations directory
labels = blood_labels, # Class labels
n_iter = 10, # Number of k-means++ iterations
annot_format = "pascal_voc", # Annotations format
seed = 55, # Random seed
centroid_fun = mean # Centroid function
)## label n
## 1 Platelets 361
## 2 RBC 4153
## 3 WBC 372
blood_anchors## [[1]]
## [[1]][[1]]
## [1] 0.3552235 0.4417515
##
## [[1]][[2]]
## [1] 0.2911290 0.3292675
##
## [[1]][[3]]
## [1] 0.1971296 0.2346442
##
##
## [[2]]
## [[2]][[1]]
## [1] 0.1757463 0.1592062
##
## [[2]][[2]]
## [1] 0.1652637 0.2065506
##
## [[2]][[3]]
## [1] 0.1630269 0.2439239
##
##
## [[3]]
## [[3]][[1]]
## [1] 0.1391842 0.1769376
##
## [[3]][[2]]
## [1] 0.1245985 0.2258089
##
## [[3]][[3]]
## [1] 0.06237392 0.08062560
Build YOLOv3 model (you can load YOLOv3
Darknet weights trained on COCO
dataset. Download pre-trained weights
from here):
blood_yolo <- yolo3(
net_h = net_h, # Input image height
net_w = net_w, # Input image width
grayscale = FALSE, # Should images be loaded as grayscale or RGB
n_class = n_class, # Number of object classes (80 for COCO dataset)
anchors = blood_anchors # Anchor boxes
)
blood_yolo %>% load_darknet_weights(here("development/yolov3.weights")) # Optional
blood_yolo## Model
## Model: "yolo3"
## ________________________________________________________________________________
## Layer (type) Output Shape Param # Connected to
## ================================================================================
## input_img (InputLayer) [(None, 416, 416, 0
## ________________________________________________________________________________
## darknet53 (Functional) [(None, None, Non 40620640 input_img[0][0]
## ________________________________________________________________________________
## yolo3_conv1 (Functional) (None, 13, 13, 51 11024384 darknet53[0][2]
## ________________________________________________________________________________
## yolo3_conv2 (Functional) (None, 26, 26, 25 2957312 yolo3_conv1[0][0]
## darknet53[0][1]
## ________________________________________________________________________________
## yolo3_conv3 (Functional) (None, 52, 52, 12 741376 yolo3_conv2[0][0]
## darknet53[0][0]
## ________________________________________________________________________________
## grid1 (Functional) (None, 13, 13, 3, 4747288 yolo3_conv1[0][0]
## ________________________________________________________________________________
## grid2 (Functional) (None, 26, 26, 3, 1194008 yolo3_conv2[0][0]
## ________________________________________________________________________________
## grid3 (Functional) (None, 52, 52, 3, 302104 yolo3_conv3[0][0]
## ================================================================================
## Total params: 61,587,112
## Trainable params: 61,534,504
## Non-trainable params: 52,608
## ________________________________________________________________________________
Compile the model with correct loss and metrics:
blood_yolo %>% compile(
optimizer = optimizer_adam(lr = 1e-5),
loss = yolo3_loss(blood_anchors, n_class = n_class),
metrics = yolo3_metrics(blood_anchors, n_class = n_class)
)Create data generators:
train_blood_yolo_generator <- yolo3_generator(
annot_path = file.path(BCCD_path, "train", "Annotations/"),
images_path = file.path(BCCD_path, "train", "JPEGImages/"),
net_h = net_h,
net_w = net_w,
batch_size = 8,
shuffle = TRUE,
labels = blood_labels
)## 291 images with corresponding annotations detected!
## Set 'steps_per_epoch' to: 37
valid_blood_yolo_generator <- yolo3_generator(
annot_path = file.path(BCCD_path, "valid", "Annotations/"),
images_path = file.path(BCCD_path, "valid", "JPEGImages/"),
net_h = net_h,
net_w = net_w,
batch_size = 8,
shuffle = TRUE,
labels = blood_labels
)## 69 images with corresponding annotations detected!
## Set 'steps_per_epoch' to: 9
Fit the model:
history <- blood_yolo %>%
yolo3_fit_generator(
generator = train_blood_yolo_generator,
epochs = 5,
steps_per_epoch = 37,
validation_generator = valid_blood_yolo_generator,
validation_steps_per_epoch = 9,
model_filepath = "development/BCCD/blood_w.hdf5",
save_best_only = TRUE,
monitor = "val_loss"
)Predict on new images:
blood_yolo <- yolo3(
net_h = net_h,
net_w = net_w,
grayscale = FALSE,
n_class = n_class,
anchors = blood_anchors
)
blood_yolo %>% load_model_weights_hdf5(here("development/BCCD/blood_w.hdf5"))
test_blood_yolo_generator <- yolo3_generator(
annot_path = file.path(BCCD_path, "test", "Annotations/"),
images_path = file.path(BCCD_path, "test", "JPEGImages/"),
net_h = net_h,
net_w = net_w,
batch_size = 4,
shuffle = FALSE,
labels = blood_labels
)## 4 images with corresponding annotations detected!
## Set 'steps_per_epoch' to: 1
test_preds <- predict_generator(blood_yolo, test_blood_yolo_generator, 1)
test_boxes <- get_boxes(test_preds, blood_anchors, blood_labels,
obj_threshold = 0.6)
test_boxes## [[1]]
## # A tibble: 17 x 7
## xmin ymin xmax ymax p_obj label_id label
## <dbl> <dbl> <dbl> <dbl> <dbl> <int> <chr>
## 1 0.617 0.0137 0.884 0.543 1 2 RBC
## 2 0.299 0 0.500 0.140 1.00 2 RBC
## 3 0.479 0 0.687 0.203 1.00 2 RBC
## 4 0.709 0 0.870 0.206 1.00 2 RBC
## 5 0.0199 0 0.163 0.250 1.00 2 RBC
## 6 0.0516 0.190 0.196 0.336 1 2 RBC
## 7 0.406 0.164 0.564 0.316 1.00 2 RBC
## 8 0.472 0.165 0.635 0.326 1.00 2 RBC
## 9 0.865 0.0795 1 0.411 1.00 2 RBC
## 10 0.0799 0.308 0.296 0.494 1.00 2 RBC
## 11 0.414 0.359 0.562 0.506 1.00 2 RBC
## 12 0.819 0.328 1 0.563 1.00 2 RBC
## 13 0.482 0.395 0.623 0.544 1.00 2 RBC
## 14 0.250 0.464 0.420 0.679 1 2 RBC
## 15 0.868 0.500 1 0.704 1.00 2 RBC
## 16 0.0584 0.531 0.290 0.821 1.00 3 WBC
## 17 0.440 0.584 0.746 0.906 1.00 3 WBC
##
## [[2]]
## # A tibble: 16 x 7
## xmin ymin xmax ymax p_obj label_id label
## <dbl> <dbl> <dbl> <dbl> <dbl> <int> <chr>
## 1 0.133 0 0.409 0.287 1.00 2 RBC
## 2 0.427 0.275 0.704 0.779 1.00 2 RBC
## 3 0.831 0.104 0.970 0.285 1.00 2 RBC
## 4 0.489 0.278 0.680 0.461 1.00 2 RBC
## 5 0.00549 0.279 0.106 0.509 1.00 2 RBC
## 6 0.579 0.306 0.757 0.496 1.00 2 RBC
## 7 0.0502 0.341 0.204 0.572 1.00 2 RBC
## 8 0.769 0.436 0.945 0.592 1.00 2 RBC
## 9 0.360 0.536 0.510 0.683 1.00 2 RBC
## 10 0.239 0.595 0.436 0.771 1.00 2 RBC
## 11 0.626 0.678 0.812 0.878 1.00 2 RBC
## 12 0.700 0.740 0.890 0.939 1.00 2 RBC
## 13 0.805 0.715 0.985 0.915 1.00 2 RBC
## 14 0.191 0.692 0.389 1 1.00 2 RBC
## 15 0.347 0.756 0.544 1 1.00 2 RBC
## 16 0.217 0.252 0.442 0.507 1.00 3 WBC
##
## [[3]]
## # A tibble: 13 x 7
## xmin ymin xmax ymax p_obj label_id label
## <dbl> <dbl> <dbl> <dbl> <dbl> <int> <chr>
## 1 0.593 0.106 0.838 0.617 0.642 2 RBC
## 2 0.212 0.577 0.434 1 0.660 2 RBC
## 3 0.280 0 0.453 0.236 1.00 2 RBC
## 4 0.637 0 0.801 0.237 1.00 2 RBC
## 5 0.780 0.184 0.972 0.399 0.996 2 RBC
## 6 0.588 0.263 0.773 0.481 0.809 2 RBC
## 7 0 0.342 0.214 0.514 0.931 2 RBC
## 8 0.836 0.468 1 0.661 1.00 2 RBC
## 9 0.778 0.686 0.969 0.873 1.00 2 RBC
## 10 0.441 0.770 0.624 0.958 0.939 2 RBC
## 11 0.610 0.783 0.806 0.969 1.00 2 RBC
## 12 0.257 0.195 0.592 0.523 0.999 3 WBC
## 13 0.225 0.259 0.512 0.559 0.999 3 WBC
##
## [[4]]
## # A tibble: 19 x 7
## xmin ymin xmax ymax p_obj label_id label
## <dbl> <dbl> <dbl> <dbl> <dbl> <int> <chr>
## 1 0.0478 0.239 0.219 0.336 0.982 1 Platelets
## 2 0.185 0.512 0.305 0.685 0.978 1 Platelets
## 3 0.318 0.578 0.496 0.689 0.998 1 Platelets
## 4 0 0.510 0.0972 0.777 1.00 1 Platelets
## 5 0.158 0 0.343 0.281 0.947 2 RBC
## 6 0.874 0.0645 1 0.235 0.962 2 RBC
## 7 0.372 0.151 0.567 0.325 1.00 2 RBC
## 8 0.549 0.155 0.709 0.478 1.00 2 RBC
## 9 0.445 0.251 0.607 0.566 0.984 2 RBC
## 10 0.458 0.339 0.639 0.514 0.988 2 RBC
## 11 0.00327 0.466 0.193 0.642 1.00 2 RBC
## 12 0.463 0.562 0.642 0.789 1.00 2 RBC
## 13 0.178 0.671 0.331 0.847 0.997 2 RBC
## 14 0.148 0.671 0.370 0.885 0.975 2 RBC
## 15 0.358 0.696 0.546 1 0.994 2 RBC
## 16 0.535 0.806 0.710 0.969 1.00 2 RBC
## 17 0.692 0.705 0.892 1 0.999 2 RBC
## 18 0.837 0.783 1 1 0.994 2 RBC
## 19 0.615 0.408 0.904 0.748 1.00 3 WBC
Plot / save images with predicted bounding boxes:
plot_boxes(
images_paths = list.files(file.path(BCCD_path, "test", "JPEGImages/"), full.names = TRUE),
boxes = test_boxes,
labels = blood_labels,
save_dir = BCCD_path)
