Add an article on segmentation with fcn_resnet model

NEWS.md

@@ -2,6 +2,7 @@
 
 ## New features
 
+* Added segmentation article on `model_fcn_resnet50()` with two images (@DerrickUnleashed, #281).
 * Added collection dataset catalog with `search_collection()`, `get_collection_catalog()`, and `list_collection_datasets()` functions for discovering and exploring collections (#271, @ANAMASGARD).
 * Added `target_transform_coco_masks()` and `target_transform_trimap_masks()` transformation functions for explicit segmentation mask generation (@ANAMASGARD).
 
@@ -12,6 +13,7 @@
 
 ## New datasets
 
+* Added `vggface2_dataset()` for loading the VGGFace2 dataset (@DerrickUnleashed, #238).
 * Added `moth` dataset to `rf100_biology_collection()` and `currency` and `wine_label` to `rf100_document_collection()` (#274).
 
 ## Bug fixes and improvements
@@ -33,7 +35,6 @@ to non-vectorized `transform_` operations (#264)
   and `rf100_underwater_collection()` . Those are collection of datasets from RoboFlow 100 under the same 
   thematic, for a total of 35 datasets (@koshtiakanksha, @cregouby, #239).
 * Added `rf100_peixos_segmentation_dataset()`.  (@koshtiakanksha, @cregouby, #250).
-* Added `vggface2_dataset()` for loading the VGGFace2 dataset (@DerrickUnleashed, #238).
 
 ## New models
 

R/dataset-vggface2.R

@@ -32,7 +32,7 @@
 #' ds$classes[item$y]  # list(name=..., gender=...)
 #' }
 #'
-#' @family segmentation_dataset
+#' @family classification_dataset
 #' @export
 vggface2_dataset <- torch::dataset(
   name = "vggface2",

R/vision_utils.R

@@ -346,7 +346,7 @@ draw_segmentation_masks.torch_tensor <- function(x,
   if (masks$dtype != torch::torch_bool() && masks$dtype != torch::torch_float() ) {
     type_error("`masks` is expected to be of dtype torch_bool() or torch_float()")
   }
-  if (any(masks$shape[2:3] != img_to_draw$shape[2:3])) {
+  if (any(masks$shape[-2:-1] != img_to_draw$shape[-2:-1])) {
     value_error("`masks` and `image` must have the same height and width")
   }
   # if mask is a model inference output, we need to convert float mask to boolean mask

_pkgdown.yml

@@ -27,6 +27,8 @@ navbar:
         href: articles/examples/style-transfer.html
       - text: texture-nca
         href: articles/examples/texture-nca.html
+      - text: fcnresnet
+        href: articles/examples/fcnresnet.html
 
 reference:
 - title: Transforms

tests/testthat/test-dataset-coco.R

@@ -16,7 +16,7 @@ test_that("coco_detection_dataset handles missing files gracefully", {
 })
 
 test_that("coco_detection_dataset loads a single example correctly", {
-  skip_if(Sys.getenv("TEST_LARGE_DATASETS", unset = 0) < 1,
+  skip_if(Sys.getenv("TEST_LARGE_DATASETS", unset = 0) != 1,
         "Skipping test: set TEST_LARGE_DATASETS=1 to enable tests requiring large downloads.")
 
   ds <- coco_detection_dataset(root = tmp, train = FALSE, year = "2017", download = TRUE)
@@ -47,7 +47,7 @@ test_that("coco_detection_dataset loads a single example correctly", {
 })
 
 test_that("coco_ dataset loads a single segmentation example correctly", {
-  skip_if(Sys.getenv("TEST_LARGE_DATASETS", unset = 0) < 1,
+  skip_if(Sys.getenv("TEST_LARGE_DATASETS", unset = 0) != 1,
         "Skipping test: set TEST_LARGE_DATASETS=1 to enable tests requiring large downloads.")
 
   ds <- coco_detection_dataset(root = tmp, train = FALSE, year = "2017", download = TRUE,
@@ -71,7 +71,7 @@ test_that("coco_ dataset loads a single segmentation example correctly", {
 })
 
 test_that("coco_detection_dataset batches correctly using dataloader", {
-  skip_if(Sys.getenv("TEST_LARGE_DATASETS", unset = 0) < 1,
+  skip_if(Sys.getenv("TEST_LARGE_DATASETS", unset = 0) != 1,
         "Skipping test: set TEST_LARGE_DATASETS=1 to enable tests requiring large downloads.")
 
 
@@ -99,8 +99,8 @@ test_that("coco_caption_dataset handles missing files gracefully", {
 })
 
 test_that("coco_caption_dataset loads a single example correctly", {
-  skip_if(Sys.getenv("TEST_LARGE_DATASETS", unset = 0) < 2,
-        "Skipping test: set TEST_LARGE_DATASETS=2 to enable tests requiring huge downloads.")
+  skip_if(Sys.getenv("TEST_HUGE_DATASETS", unset = 0) != 1,
+        "Skipping test: set TEST_HUGE_DATASETS=1 to enable tests requiring huge downloads.")
 
   ds <- coco_caption_dataset(root = tmp, train = FALSE, download = TRUE)
 
@@ -120,8 +120,8 @@ test_that("coco_caption_dataset loads a single example correctly", {
 })
 
 test_that("coco_caption_dataset batches correctly using dataloader", {
-  skip_if(Sys.getenv("TEST_LARGE_DATASETS", unset = 0) < 2,
-        "Skipping test: set TEST_LARGE_DATASETS=2 to enable tests requiring huge downloads.")
+  skip_if(Sys.getenv("TEST_HUGE_DATASETS", unset = 0) != 1,
+        "Skipping test: set TEST_HUGE_DATASETS=1 to enable tests requiring huge downloads.")
 
   ds <- coco_caption_dataset(root = tmp, train = FALSE, download = TRUE)
 

vignettes/examples/fcnresnet.R

@@ -0,0 +1,69 @@
+# Loading Images ---------------------------------------------------
+library(torchvision)
+library(torch)
+
+url1 <- "https://raw.githubusercontent.com/pytorch/vision/main/gallery/assets/dog1.jpg"
+url2 <- "https://raw.githubusercontent.com/pytorch/vision/main/gallery/assets/dog2.jpg"
+
+dog1 <- magick_loader(url1) |> transform_to_tensor()
+dog2 <- magick_loader(url2) |> transform_to_tensor()
+
+
+# Visualizing a grid of images -------------------------------------
+
+
+dogs <- torch_stack(list(dog1, dog2))
+grid <- vision_make_grid(dogs, scale = TRUE, num_rows = 2)
+tensor_image_browse(grid)
+
+
+# Preprocessing the data -------------------------------------
+
+
+norm_mean <- c(0.485, 0.456, 0.406)
+norm_std  <- c(0.229, 0.224, 0.225)
+
+dog1_prep <- dog1 |>
+  transform_resize(c(520,520)) |>
+  transform_normalize(mean = norm_mean, std = norm_std)
+dog2_prep <- dog2 |>
+  transform_resize(c(520,520)) |>
+  transform_normalize(mean = norm_mean, std = norm_std)
+
+# make batch (2,3,520,520)
+dog_batch <- torch_stack(list(dog1_prep, dog2_prep))
+
+
+# Loading Model -------------------------------------
+
+
+model <- model_fcn_resnet50(pretrained = TRUE)
+model$eval()
+
+# run model
+output <- model(dog_batch)
+
+
+# Processing the Output ------------------------------
+
+mask <- output$out
+mask$shape
+mask$dtype
+
+# Visualizing the Output ------------------------------
+
+
+segmented1 <- draw_segmentation_masks(
+  dog1 |> transform_resize(c(520,520)),
+  masks = mask[1,, ],
+  alpha = 0.5
+)
+
+segmented2 <- draw_segmentation_masks(
+  dog2 |> transform_resize(c(520,520)),
+  masks = mask[2,, ],
+  alpha = 0.5
+)
+
+tensor_image_browse(segmented1)
+tensor_image_browse(segmented2)

vignettes/examples/fcnresnet.Rmd

@@ -0,0 +1,9 @@
+---
+title: "fcnresnet"
+type: docs
+---
+
+```{r, echo = FALSE}
+knitr::opts_chunk$set(eval = TRUE)
+knitr::spin_child(paste0(rmarkdown::metadata$title, ".R"))
+```