summaryrefslogtreecommitdiff
path: root/text_recognizer/data/transforms/embed_crop.py
diff options
context:
space:
mode:
authorGustaf Rydholm <gustaf.rydholm@gmail.com>2022-09-27 23:18:20 +0200
committerGustaf Rydholm <gustaf.rydholm@gmail.com>2022-09-27 23:18:20 +0200
commit723cf87846bf7297326fc82973a1e148af317638 (patch)
tree0bd49e42a3ca8c2c9e16472355b7f262cef27e05 /text_recognizer/data/transforms/embed_crop.py
parent2dcbd13eb311d0a4b31e086a53d2afc55d50e0b5 (diff)
Remove transforms
Diffstat (limited to 'text_recognizer/data/transforms/embed_crop.py')
-rw-r--r--text_recognizer/data/transforms/embed_crop.py37
1 files changed, 0 insertions, 37 deletions
diff --git a/text_recognizer/data/transforms/embed_crop.py b/text_recognizer/data/transforms/embed_crop.py
deleted file mode 100644
index 7421d0e..0000000
--- a/text_recognizer/data/transforms/embed_crop.py
+++ /dev/null
@@ -1,37 +0,0 @@
-"""Transforms for PyTorch datasets."""
-import random
-
-from PIL import Image
-
-
-class EmbedCrop:
-
- IMAGE_HEIGHT = 56
- IMAGE_WIDTH = 1024
-
- def __init__(self, augment: bool) -> None:
- self.augment = augment
-
- def __call__(self, crop: Image) -> Image:
- # Crop is PIL.Image of dtype="L" (so value range is [0, 255])
- image = Image.new("L", (self.IMAGE_WIDTH, self.IMAGE_HEIGHT))
-
- # Resize crop.
- crop_width, crop_height = crop.size
- new_crop_height = self.IMAGE_HEIGHT
- new_crop_width = int(new_crop_height * (crop_width / crop_height))
-
- if self.augment:
- # Add random stretching
- new_crop_width = int(new_crop_width * random.uniform(0.9, 1.1))
- new_crop_width = min(new_crop_width, self.IMAGE_WIDTH)
- crop_resized = crop.resize(
- (new_crop_width, new_crop_height), resample=Image.BILINEAR
- )
-
- # Embed in image
- x = min(28, self.IMAGE_WIDTH - new_crop_width)
- y = self.IMAGE_HEIGHT - new_crop_height
- image.paste(crop_resized, (x, y))
-
- return image