Feat: add item_transform_pad for detection and segmentation items

[DerrickUnleashed]

Added item_transform_pad for detection item, dataset and segmentation item, dataset

Closes #352

[DerrickUnleashed]

url <- "https://upload.wikimedia.org/wikipedia/commons/b/b6/Felis_catus-cat_on_snow.jpg"

# ========== 1. DETECTION ITEM ==========
img <- base_loader(url) |> transform_to_tensor()
boxes <- torch_tensor(matrix(c(600, 200, 2880, 1860), ncol = 4), dtype = torch_float32())
det_item <- list(x = img, y = list(boxes = boxes, labels = "CAT"))
class(det_item) <- c("image_with_bounding_box", "list")

det_padded <- item_transform_pad(det_item, padding = 100)

p1 <- draw_bounding_boxes(det_item, colors = "blue", width = 10)$to(torch_float())$div(255)
p2 <- draw_bounding_boxes(det_padded, colors = "red", width = 10)$to(torch_float())$div(255)
p1 <- transform_resize(p1, c(300, 300))
p2 <- transform_resize(p2, c(300, 300))
g1 <- vision_make_grid(torch_stack(list(p1, p2)), scale = TRUE)
tensor_image_browse(g1)

# ========== 2. DETECTION DATASET ==========
ds_det <- pascal_detection_dataset(year = "2007", split = "trainval",
                                   transform = transform_to_tensor, download = TRUE)
orig_det <- ds_det[1]
ds_det_padded <- item_transform_pad(ds_det, padding = c(50, 50, 50, 50))
pad_det <- ds_det_padded[1]

p3 <- draw_bounding_boxes(orig_det, colors = "blue", width = 5)$to(torch_float())$div(255)
p4 <- draw_bounding_boxes(pad_det, colors = "red", width = 5)$to(torch_float())$div(255)
p3 <- transform_resize(p3, c(300, 300))
p4 <- transform_resize(p4, c(300, 300))
g2 <- vision_make_grid(torch_stack(list(p3, p4)), scale = TRUE)
tensor_image_browse(g2)

# ========== 3. SEGMENTATION ITEM ==========
img2 <- base_loader(url) |> transform_to_tensor()
h <- img2$shape[2]; w <- img2$shape[3]

mask1 <- torch_zeros(h, w, dtype = torch_bool())
mask1[, 1:150] <- TRUE

mask2 <- torch_zeros(h, w, dtype = torch_bool())
mask2[, 350:500] <- TRUE

masks <- torch_stack(list(mask1, mask2))
seg_item <- list(x = img2, y = list(masks = masks, labels = torch_tensor(c(1L, 2L)),
                                    image_height = h, image_width = w))
class(seg_item) <- c("image_with_segmentation_mask", "list")

seg_padded <- item_transform_pad(seg_item, padding = 100)

p5 <- draw_segmentation_masks(seg_item, alpha = 0.5, colors = c("red", "blue"))$to(torch_float())$div(255)
p6 <- draw_segmentation_masks(seg_padded, alpha = 0.5, colors = c("red", "blue"))$to(torch_float())$div(255)
p5 <- transform_resize(p5, c(300, 300))
p6 <- transform_resize(p6, c(300, 300))
g3 <- vision_make_grid(torch_stack(list(p5, p6)), scale = TRUE)
tensor_image_browse(g3)

# ========== 4. SEGMENTATION DATASET ==========
ds_seg <- pascal_segmentation_dataset(year = "2007", split = "trainval",
                                      transform = transform_to_tensor, download = TRUE)
orig_seg <- ds_seg[1]
ds_seg_padded <- item_transform_pad(ds_seg, padding = c(50, 50, 50, 50))
pad_seg <- ds_seg_padded[1]

p7 <- draw_segmentation_masks(orig_seg, alpha = 0.5)$to(torch_float())$div(255)
p8 <- draw_segmentation_masks(pad_seg, alpha = 0.5)$to(torch_float())$div(255)
p7 <- transform_resize(p7, c(300, 300))
p8 <- transform_resize(p8, c(300, 300))
g4 <- vision_make_grid(torch_stack(list(p7, p8)), scale = TRUE)
tensor_image_browse(g4)

DETECTION ITEM

DETECTION DATASET

SEGMENTATION ITEM

SEGMENTATION DATASET

[cregouby]

Praise Thanks for this !

[cregouby]

todo This cannot be part of #373 as it is already in #370 and #371, and thus prevents the independant merging of this P.R.
suggestion you should derive a new branch from main, not from the previous branch you where working on to keep them independant.

[DerrickUnleashed]

Ohh yes that would be better from the next PR onwards I'll cut branches from main

[cregouby]

todo This cannot be part of #373 as it is already in #370 and #371, and thus prevents the independant merging of this P.R.

[cregouby]

todo This cannot be part of #373 as it is already in #370 and #371, and thus prevents the independant merging of this P.R.

[cregouby]

todo This cannot be part of #373 as it is already in #370 and #371, and thus prevents the independant merging of this P.R.

[DerrickUnleashed]

why I cut from existing branches ie (hflip cut from main, vflip cut from hflip, center crop cut from vflip and so on) was because since im making changes in the same file it'll be easier to fix merge conflicts but however it does have this issue of not being able to review PRs simultaneously which I didn't consider sorry about that, can we merge #371 and then on this PR I'll make sure that from the next PR onwards I cut from main