animl 3.1.1

.Rbuildignore

@@ -6,4 +6,5 @@
 ^.Rhistory
 ^.*\.Rproj$
 ^\.Rproj\.user$
-^R/old/*
+^R/old/*
+^.git

.gitignore

@@ -1,3 +1,4 @@
+.RData
 .Rhistory
 .Rproj.user
 inst/doc

DESCRIPTION

@@ -1,6 +1,6 @@
 Package: animl
 Title: A Collection of ML Tools for Conservation Research
-Version: 3.0.0
+Version: 3.1.1
 Authors@R: c(person(given="Kyra", family="Swanson",email="tswanson@sdzwa.org", role = c("aut", "cre"),
            comment = c(ORCID = "0000-0002-1496-3217")), person(given="Mathias",family="Tobler",role = "aut"))
 Description: Functions required to classify subjects within camera trap field data. The package can handle both images and videos. The authors recommend a two-step approach using Microsoft's 'MegaDector' model and then a second model trained on the classes of interest.
@@ -12,7 +12,6 @@ Imports:
     methods,
     pbapply,
     reticulate,
-    parallel,
     stats,
 Depends: 
     R (>= 4.0.0)

NAMESPACE

@@ -13,19 +13,23 @@ export(delete_pyenv)
 export(detect)
 export(download_model)
 export(euclidean_squared_distance)
+export(export_camtrapR)
+export(export_coco)
 export(export_folders)
 export(export_megadetector)
 export(export_timelapse)
 export(extract_frames)
 export(extract_miew_embeddings)
 export(get_animals)
 export(get_empty)
+export(get_frame_as_image)
 export(list_models)
 export(load_animl_py)
 export(load_class_list)
 export(load_classifier)
 export(load_data)
 export(load_detector)
+export(load_json)
 export(load_miew)
 export(parse_detections)
 export(plot_all_bounding_boxes)
@@ -34,10 +38,12 @@ export(plot_from_file)
 export(remove_diagonal)
 export(remove_link)
 export(save_classifier)
+export(save_json)
 export(sequence_classification)
 export(single_classification)
 export(test_main)
 export(train_main)
 export(train_val_test)
+export(update_animl_py)
 export(update_labels_from_folders)
 importFrom(methods,is)

R/classification.R

@@ -1,20 +1,22 @@
-#' Load a Classifier Model with animl-py
+#' Load a Classifier Model and Class_list
 #'
 #' @param model_path path to model
-#' @param len_classes path to class list
+#' @param classes path to class list or loaded class list
 #' @param device send model to the specified device
 #' @param architecture model architecture
 #'
-#' @return classifier model
+#' @return classifier model, class list
 #' @export
 #'
 #' @examples
 #' \dontrun{
 #' classes <- load_class_list('sdzwa_andes_v1_classes.csv')
 #' andes <- load_classifier('andes_v1.pt', nrow(classes))}
-load_classifier <- function(model_path, len_classes, device=NULL, architecture="CTL"){
+load_classifier <- function(model_path, classes, device=NULL, architecture="CTL"){
   animl_py <- get("animl_py", envir = parent.env(environment()))
-  animl_py$load_classifier(model_path, as.integer(len_classes), device=device, architecture=architecture)
+
+  if(is.numeric(classes)){ classes = as.integer(classes)}
+  animl_py$load_classifier(model_path, classes, device=device, architecture=architecture)
 }
 
 
@@ -47,38 +49,43 @@ save_classifier <- function(model, out_dir, epoch, stats, optimizer=NULL, schedu
 #' @examples
 #' \dontrun{classes <- load_class_list('andes_classes.csv')}
 load_class_list <- function(classlist_file){
-  read.csv(classlist_file)
+  utils::read.csv(classlist_file)
 }
 
 
 #' Infer Species for Given Detections
 #'
 #' @param model loaded classifier model
 #' @param detections manifest of animal detections
-#' @param device send model to the specified device
-#' @param out_file path to csv to save results to
+#' @param resize_width image width input size
+#' @param resize_height image height input size
 #' @param file_col column in manifest containing file paths
 #' @param crop use bbox to crop images before feeding into model
 #' @param normalize normalize the tensor before inference
-#' @param resize_width image width input size
-#' @param resize_height image height input size
 #' @param batch_size batch size for generator 
-#' @param num_workers number of processes 
+#' @param num_workers number of processes
+#' @param device send model to the specified device
+#' @param out_file path to csv to save results to
 #'
 #' @return detection manifest with added prediction and confidence columns
 #' @export
 #'
 #' @examples
 #' \dontrun{animals <- classify(classifier, animals, file_col='filepath')}
-  classify <- function(model, detections, device=NULL, out_file=NULL,
-                       file_col='frame', crop=TRUE, normalize=TRUE,
+  classify <- function(model, detections, 
                        resize_width=480, resize_height=480,
-                       batch_size=1, num_workers=1){
+                       file_col='filepath', crop=TRUE, normalize=TRUE,
+                       batch_size=1, num_workers=1,
+                       device=NULL, out_file=NULL){
+
     animl_py <- get("animl_py", envir = parent.env(environment()))
-    animl_py$classify(model, detections, device=device, out_file=out_file,
-                      file_col=file_col, crop=crop, normalize=normalize, 
-                      resize_width=as.integer(resize_width), resize_height=as.integer(resize_height),
-                      batch_size=as.integer(batch_size), num_workers=as.integer(num_workers))
+    animl_py$classify(model, detections,
+                      resize_width=as.integer(resize_width),
+                      resize_height=as.integer(resize_height), 
+                      file_col=file_col, crop=crop, normalize=normalize,
+                      batch_size=as.integer(batch_size),
+                      num_workers=as.integer(num_workers),
+                      device=device, out_file=out_file)
 }
 
 

R/detection.R

@@ -1,14 +1,15 @@
 #' Load an Object Detector
 #'
 #' @param model_path path to detector model file
-#' @param model_type type of model expected ie "MDV5", "MDV6", "YOLO"
+#' @param model_type type of model expected ie "MDV5", "MDV6", "YOLO", "ONNX"
 #' @param device specify to run on cpu or gpu
 #'
-#' @return megadetector object
+#' @return detector object
 #' @export
 #'
 #' @examples
-#' \dontrun{md_py <- megadetector("/mnt/machinelearning/megaDetector/md_v5a.0.0.pt", model_type='mdv5', device = 'cuda:0')}
+#' \dontrun{md_py <- megadetector("/mnt/machinelearning/megaDetector/md_v5a.0.0.pt", 
+#'                                model_type='mdv5', device='cuda:0')}
 load_detector <- function(model_path, model_type, device=NULL){
   # first check if animl-py is loaded
   animl_py$load_detector(model_path, model_type=model_type, device=device)
@@ -36,7 +37,7 @@ load_detector <- function(model_path, model_type, device=NULL){
 #' @examples
 #' \dontrun{mdres <- detect(md_py, allframes$Frame, 1280, 960, device='cpu')}
 detect <- function(detector, image_file_names, resize_width, resize_height,
-                   letterbox=TRUE, confidence_threshold=0.1, file_col='frame',
+                   letterbox=TRUE, confidence_threshold=0.1, file_col='filepath',
                    batch_size=1, num_workers=1, device=NULL,
                    checkpoint_path=NULL, checkpoint_frequency=-1){
   animl_py <- get("animl_py", envir = parent.env(environment()))
@@ -50,11 +51,11 @@ detect <- function(detector, image_file_names, resize_width, resize_height,
 }
 
 
-#' parse MD results into a simple dataframe
+#' Parse MD results into a simple dataframe
 #'
 #' @param results json output from megadetector
-#' @param manifest dataframe containing all frames
-#' @param out_file path to save dataframe
+#' @param manifest optional dataframe containing all frames
+#' @param out_file optional path to save dataframe
 #' @param threshold confidence threshold to include bbox
 #' @param file_col column in manifest that refers to file paths 
 #' 
@@ -65,7 +66,7 @@ detect <- function(detector, image_file_names, resize_width, resize_height,
 #' \dontrun{
 #' mdresults <- parseMD(mdres)
 #' }
-parse_detections <- function(results, manifest=NULL, out_file=NULL, threshold=0, file_col="frame") {
+parse_detections <- function(results, manifest=NULL, out_file=NULL, threshold=0, file_col="filepath") {
   animl_py <- get("animl_py", envir = parent.env(environment()))
   animl_py$parse_detections(results, manifest=manifest, out_file=out_file,
                             threshold=threshold, file_col=file_col)

R/export.R

@@ -48,21 +48,21 @@ remove_link <- function(manifest, link_col='link'){
 #' Udate Results from File Browser
 #'
 #' @param manifest dataframe containing file data and predictions
-#' @param link_dir directory to sort files into
+#' @param export_dir directory to sort files into
 #' @param unique_name column name indicating a unique file name for each row
 #'
 #' @return dataframe with new "Species" column that contains the verified species
 #' @export
 #'
 #' @examples
 #' \dontrun{
-#' results <- update_labels_from_folders(manifest, link_dir)
+#' results <- update_labels_from_folders(manifest, export_dir)
 #' }
-update_labels_from_folders <- function(manifest, link_dir, unique_name='uniquename'){
-  if (!dir.exists(link_dir)) {stop("The given directory does not exist.")}
+update_labels_from_folders <- function(manifest, export_dir, unique_name='uniquename'){
+  if (!dir.exists(export_dir)) {stop("The given directory does not exist.")}
   if (!unique_name %in% names(manifest)) {stop("Manifest does not have unique names, cannot match to sorted directories.")}
 
-  FilePath <- list.files(link_dir, recursive = TRUE, include.dirs = TRUE)
+  FilePath <- list.files(export_dir, recursive = TRUE, include.dirs = TRUE)
   files <- data.frame(FilePath)
 
   files[unique_name] <- sapply(files$FilePath,function(x)strsplit(x,"/")[[1]][2])
@@ -73,36 +73,88 @@ update_labels_from_folders <- function(manifest, link_dir, unique_name='uniquena
 }
 
 
-#' Converts the .csv file to the MD-formatted .json file.
+#' Converts the .csv file to a COCO-formatted .json file.
 #'
 #' @param manifest dataframe containing images and associated detections
-#' @param output_file path to save the MD formatted file
-#' @param detector name of the detector model used
+#' @param class_list  dataframe containing class names and their corresponding IDs
+#' @param out_file path to save the formatted file
+#' @param info info section of COCO file, named list
+#' @param licenses licenses section of COCO file, array
 #'
-#' @return None
+#' @return coco formated json
 #' @export
 #'
 #' @examples
 #' \dontrun{export_megadetector(manifest, output_file= 'results.json', detector='MDv6')}
-export_megadetector <- function(manifest, output_file=NULL, detector='MegaDetector v5a'){
+export_coco <- function(manifest, class_list, out_file, info=NULL, licenses=NULL){
   animl_py <- get("animl_py", envir = parent.env(environment()))
-  animl_py$export_megadetector(manifest, output_file=output_file, detector=detector)
+  animl_py$export_coco(manifest, class_list, out_file, info=info, licenses=licenses)
+}
+
+
+
+#' Export data into sorted folders organized by station
+#'
+#' @param manifest dataframe containing images and associated predictions
+#' @param out_dir directory to export sorted images
+#' @param out_file if provided, save the manifest to this file
+#' @param label_col column containing species labels
+#' @param file_col column containing source paths
+#' @param station_col column containing station names
+#' @param unique_name column containing unique file name
+#' @param copy if true, hard copy
+#'
+#' @returns manifest with link column
+#' @export
+#'
+#' @examples
+#' \dontrun{manifest <- export_camtrapR(manifest, out_dir, out_file=NULL, label_col='prediction',
+#'                                      file_col="filepath", station_col='station', 
+#'                                      unique_name='uniquename', copy=FALSE)}
+export_camtrapR <- function(manifest, out_dir, out_file=NULL, label_col='prediction',
+                            file_col="filepath", station_col='station', 
+                            unique_name='uniquename', copy=FALSE){
+  animl_py <- get("animl_py", envir = parent.env(environment()))
+  animl_py$export_camtrapR(manifest, out_dir, out_file=out_file, label_col=label_col,
+                           file_col=file_col, station_col=station_col,
+                           unique_name=unique_name, copy=copy)
 }
 
 
 #' Converts the Manifests to a csv file that contains columns needed for TimeLapse conversion in later step
 #'
-#' @param animals a DataFrame that has entries of anuimal classification
-#' @param empty a DataFrame that has detection of non-animal objects in images
-#' @param imagedir location of root directory where all images are stored (can contain subdirectories)
+#' @param results a DataFrame that has entries of anuimal classification
+#' @param image_dir location of root directory where all images are stored (can contain subdirectories)
 #' @param only_animal A bool that confirms whether we want only animal detctions or all
 #'
 #' @returns animals.csv, non-anim.csv, csv_loc
 #' @export 
 #'
 #' @examples
 #' \dontrun{export_timelapse(animals, empty, '/path/to/images/')}
-export_timelapse <- function(animals, empty, imagedir, only_animal=TRUE){
+export_timelapse <- function(results, image_dir, only_animal=TRUE){
+  animl_py <- get("animl_py", envir = parent.env(environment()))
+  animl_py$export_timelapse(results, image_dir, only_animal=only_animal)
+}
+
+
+#' Converts the .csv file to the MD-formatted .json file.
+#'
+#' @param manifest dataframe containing images and associated detections
+#' @param out_file path to save the MD formatted file
+#' @param detector name of the detector model used
+#' @param prompt ask user to overwrite existing file
+#'
+#' @return None
+#' @export
+#'
+#' @examples
+#' \dontrun{export_megadetector(manifest, output_file= 'results.json', detector='MDv6')}
+export_megadetector <- function(manifest, out_file=NULL, 
+                                detector='MegaDetector v5a', prompt=TRUE){
   animl_py <- get("animl_py", envir = parent.env(environment()))
-  animl_py$export_timelapse(animals, empty, imagedir, only_animal=only_animal)
+  animl_py$export_megadetector(manifest, out_file=out_file, 
+                               detector=detector, prompt=prompt)
 }
+
+