Fix: item_transform_rotate keeps tight bounding boxes

[DerrickUnleashed]

• item_transform_rotate now rotates detection boxes around the image centre without expanding them to their enclosing AABB returns xyxyr with accumulated angle.
• draw_bounding_boxes.image_with_rotated_box no longer zigzags between multiple boxes - each rotated box is drawn as its own tight rectangle.

Closes #382

[DerrickUnleashed]

url <- "https://upload.wikimedia.org/wikipedia/commons/b/b6/Felis_catus-cat_on_snow.jpg"
angle <- 30

resize_pair <- function(a, b, size = c(600, 600)) {
  list(transform_resize(a, size), transform_resize(b, size))
}

# ========== 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_rotated <- item_transform_rotate(det_item, angle = angle)

p1 <- draw_bounding_boxes(det_item, colors = "blue", width = 10)$to(torch_float())$div(255)
p2 <- draw_bounding_boxes(det_rotated, colors = "red", width = 10)$to(torch_float())$div(255)
grid1 <- do.call(vision_make_grid, c(torch_stack(resize_pair(p1, p2)), list(scale = TRUE)))
tensor_image_browse(grid1)

# ========== 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_rotated <- item_transform_rotate(ds_det, angle = angle)
rot_det <- ds_det_rotated[1]

p3 <- draw_bounding_boxes(orig_det, colors = "blue", width = 5)$to(torch_float())$div(255)
p4 <- draw_bounding_boxes(rot_det, colors = "red", width = 5)$to(torch_float())$div(255)
grid2 <- do.call(vision_make_grid, c(torch_stack(resize_pair(p3, p4)), list(scale = TRUE)))
tensor_image_browse(grid2)

# ========== 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_rotated <- item_transform_rotate(seg_item, angle = angle)

p5 <- draw_segmentation_masks(seg_item, alpha = 0.5, colors = c("red", "blue"))$to(torch_float())$div(255)
p6 <- draw_segmentation_masks(seg_rotated, alpha = 0.5, colors = c("red", "blue"))$to(torch_float())$div(255)
grid3 <- do.call(vision_make_grid, c(torch_stack(resize_pair(p5, p6)), list(scale = TRUE)))
tensor_image_browse(grid3)

# ========== 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_rotated <- item_transform_rotate(ds_seg, angle = angle)
rot_seg <- ds_seg_rotated[1]

p7 <- draw_segmentation_masks(orig_seg, alpha = 0.5)$to(torch_float())$div(255)
p8 <- draw_segmentation_masks(rot_seg, alpha = 0.5)$to(torch_float())$div(255)
grid4 <- do.call(vision_make_grid, c(torch_stack(resize_pair(p7, p8)), list(scale = TRUE)))
tensor_image_browse(grid4)

DETECTION ITEM

DETECTION DATASET

SEGMENTATION ITEM

SEGMENTATION DATASET

[DerrickUnleashed]

After fix

library(torchvision)

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

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")

# expand = FALSE -> same canvas, box stays tight & angle preserved
det_rotated <- det_item |> item_transform_rotate(30, expand = FALSE)
# expand = TRUE -> bigger canvas, then crop back to original size
det_rotated_expand <- det_item |>
  item_transform_rotate(30, expand = TRUE) |>
  item_transform_crop(615, 300, 2000, 3000)

# draw_bounding_boxes returns uint8; convert to float [0,1] so
# vision_make_grid(scale = TRUE) doesn't choke on in-place float ops on Byte
p3 <- draw_bounding_boxes(det_item,           colors = "black",      width = 10)$to(torch_float())$div(255)
p4 <- draw_bounding_boxes(det_rotated,        colors = "darkgreen",  width = 10)$to(torch_float())$div(255)
p5 <- draw_bounding_boxes(det_rotated_expand, colors = "darkorange", width = 10)$to(torch_float())$div(255)

grid2 <- vision_make_grid(torch_stack(list(p3, p4, p5)))
tensor_image_browse(grid2)