Package 'ovml.common'

The IOU of two bounding boxes

Description

The IOU of two bounding boxes

Usage

bbox_iou(box1, box2)

Arguments

box1	numeric: 4-element box vector
box2	matrix: nx4 matrix of boxes

Value

A vector of length n with the intersection over union (IOU) value for each box pair

---

Check file SHA1 hash against expected

Description

Check file SHA1 hash against expected

Usage

check_sha1(filename, expected = NULL)

Arguments

filename	string: file to check
expected	string: expected SHA1 hash

Value

TRUE if the SHA1 hash matches the expected value, or if no expected value was provided.

---

Path to the cache directory used for model weight files and other data

Description

Path to the cache directory used for model weight files and other data

Usage

ovml_cache_dir()

Value

The path as a string

---

Class labels

Description

Class labels

Usage

ovml_class_labels(dataset = "coco")

Arguments

dataset

string: which dataset? One of "coco" (used with a variety of models) "mvb" (used with e.g. the yolov4-mvb model)

Value

A character vector of class labels

---

Conditional download utility

Description

If the file already exists in the ovml_cache_dir(), it won't be downloaded.

Usage

ovml_download_if(url, dest, expected_sha1 = NULL)

Arguments

url	string: URL of file to download
dest	string: local basename of file, if missing will be taken from the URL
expected_sha1	string: the expected SHA1 hash of the file

Value

The path to the file in the ovml_cache_dir()

---

Example images

Description

Example images

Usage

ovml_example_image(choices = 1)

Arguments

choices

integer: which image files to return? 1 - an image from a match between GKS Katowice and MKS Bedzin during the 2018/19 Polish Plus Liga 2 - the standard YOLO dog image

Value

Path to the image files

---

Preview plot of detections over image using base graphics

Description

Preview plot of detections over image using base graphics

Usage

ovml_plot(
  img,
  detections,
  line_args = list(col = "blue", lwd = 1),
  label_args = list(col = "white", cex = 0.75)
)

ovml_ggplot(
  img,
  detections,
  line_args = list(col = "blue", size = 0.75, fill = NA),
  label_args = list(col = "white", size = 2.5, fill = "blue"),
  label_geom = "label"
)

Arguments

img	string or image: filename of jpg image, or image as read by [jpeg::readJPEG()]
detections	data.frame: as returned by e.g. ovml::ovml_yolo_detect()
line_args	list: parameters passed to lines (for ovml_plot()) or [ggplot2::geom_rect()] (for ovml_ggplot())
label_args	list: parameters passed to text
label_geom	string: for ovml_ggplot(), the geom function to use for labels. Either "text" (use [ggplot2::geom_text()]) or "label" ([ggplot2::geom_label()])

Examples

## Not run: 
  ## define some demo data
  dets <- data.frame(class = rep("person", 3),
                     score = rep(0.99, 3),
                     xmin = c(829, 611, 736),
                     xmax = c(960, 733, 836),
                     ymin = c(88, 258, 213),
                     ymax = c(278, 444, 385),
                     stringsAsFactors = FALSE)
  img <- ovml_example_image(1)
  ovml_plot(img, dets, line_args = list(col = "red", lwd = 2))
  ovml_ggplot(img, dets) + ggplot2::theme_void()

## End(Not run)

---

ovml.common

Description

Support functions for other openvolley machine learning packages in volleyball analytics.

---

Process raw detections from pose detection network

Description

Process raw detections from pose detection network

Usage

process_pose_dets(
  pose,
  original_w,
  original_h,
  input_image_size,
  as = "segments",
  letterboxing = FALSE
)

Arguments

pose	matrix: pose detection network output
original_w	integer: input image width
original_h	integer: input image height
input_image_size	integer: network image size
as	string: return results as "segments" or "keypoints"
letterboxing	logical: TRUE if the input images were letterboxed to retain their original aspect ratio

Value

A data.frame

---

Rescale boxes

Description

Detection boxes are generally on a scaled image, with size according to the network configuration. This function takes boxes on those images and rescales back to the original image dimensions, optionally accounting for letterboxing.

Usage

rescale_boxes(
  bboxes,
  original_w,
  original_h,
  input_image_size,
  letterboxing = TRUE
)

Arguments

bboxes	numeric: boxes
original_w	numeric: original image width in pixels
original_h	numeric: original image height in pixels
input_image_size	numeric: the network image size in pixels
letterboxing	logical: were the input images letterboxed?

Value

Rescaled boxes

---

Internal utilities

Description

Internal utilities

Usage

str_trim(x)

is_num_scalar(x)

x %eq% y

make_divisible(x, divisor)

image_wh(im)

image_resz(im, sz, preserve_aspect = TRUE)

Arguments

x, y, divisor, im, sz, preserve_aspect

: generic input parms

---

Convert xywh format to bounding box format

Description

Convert xywh format to bounding box format

Usage

xywh2box(prediction)

Arguments

prediction

matrix: n x m x q matrix of predictions, where the first 4 columns in q are x y w h

Value

A matrix of the same size, with coordinates changed to xmin, ymin, xmax, ymax

---