Datasets:
refactor: remove useless
Browse files- facial_keypoint_detection.py +0 -94
facial_keypoint_detection.py
CHANGED
|
@@ -1,8 +1,5 @@
|
|
| 1 |
import datasets
|
| 2 |
-
import numpy as np
|
| 3 |
import pandas as pd
|
| 4 |
-
import PIL.Image
|
| 5 |
-
import PIL.ImageOps
|
| 6 |
|
| 7 |
_CITATION = """\
|
| 8 |
@InProceedings{huggingface:dataset,
|
|
@@ -27,66 +24,6 @@ _LICENSE = "cc-by-nc-nd-4.0"
|
|
| 27 |
_DATA = f"https://huggingface.co/datasets/TrainingDataPro/{_NAME}/resolve/main/data/"
|
| 28 |
|
| 29 |
|
| 30 |
-
def exif_transpose(img):
|
| 31 |
-
if not img:
|
| 32 |
-
return img
|
| 33 |
-
|
| 34 |
-
exif_orientation_tag = 274
|
| 35 |
-
|
| 36 |
-
# Check for EXIF data (only present on some files)
|
| 37 |
-
if hasattr(img, "_getexif") and isinstance(
|
| 38 |
-
img._getexif(), dict) and exif_orientation_tag in img._getexif():
|
| 39 |
-
exif_data = img._getexif()
|
| 40 |
-
orientation = exif_data[exif_orientation_tag]
|
| 41 |
-
|
| 42 |
-
# Handle EXIF Orientation
|
| 43 |
-
if orientation == 1:
|
| 44 |
-
# Normal image - nothing to do!
|
| 45 |
-
pass
|
| 46 |
-
elif orientation == 2:
|
| 47 |
-
# Mirrored left to right
|
| 48 |
-
img = img.transpose(PIL.Image.FLIP_LEFT_RIGHT)
|
| 49 |
-
elif orientation == 3:
|
| 50 |
-
# Rotated 180 degrees
|
| 51 |
-
img = img.rotate(180)
|
| 52 |
-
elif orientation == 4:
|
| 53 |
-
# Mirrored top to bottom
|
| 54 |
-
img = img.rotate(180).transpose(PIL.Image.FLIP_LEFT_RIGHT)
|
| 55 |
-
elif orientation == 5:
|
| 56 |
-
# Mirrored along top-left diagonal
|
| 57 |
-
img = img.rotate(-90,
|
| 58 |
-
expand=True).transpose(PIL.Image.FLIP_LEFT_RIGHT)
|
| 59 |
-
elif orientation == 6:
|
| 60 |
-
# Rotated 90 degrees
|
| 61 |
-
img = img.rotate(-90, expand=True)
|
| 62 |
-
elif orientation == 7:
|
| 63 |
-
# Mirrored along top-right diagonal
|
| 64 |
-
img = img.rotate(90,
|
| 65 |
-
expand=True).transpose(PIL.Image.FLIP_LEFT_RIGHT)
|
| 66 |
-
elif orientation == 8:
|
| 67 |
-
# Rotated 270 degrees
|
| 68 |
-
img = img.rotate(90, expand=True)
|
| 69 |
-
|
| 70 |
-
return img
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
def load_image_file(file, mode='RGB'):
|
| 74 |
-
# Load the image with PIL
|
| 75 |
-
img = PIL.Image.open(file)
|
| 76 |
-
|
| 77 |
-
if hasattr(PIL.ImageOps, 'exif_transpose'):
|
| 78 |
-
# Very recent versions of PIL can do exit transpose internally
|
| 79 |
-
img = PIL.ImageOps.exif_transpose(img)
|
| 80 |
-
else:
|
| 81 |
-
# Otherwise, do the exif transpose ourselves
|
| 82 |
-
img = exif_transpose(img)
|
| 83 |
-
|
| 84 |
-
img = img.convert(mode)
|
| 85 |
-
img.thumbnail((1000, 1000), PIL.Image.Resampling.LANCZOS)
|
| 86 |
-
|
| 87 |
-
return img
|
| 88 |
-
|
| 89 |
-
|
| 90 |
class FacialKeypointDetection(datasets.GeneratorBasedBuilder):
|
| 91 |
|
| 92 |
def _info(self):
|
|
@@ -103,13 +40,9 @@ class FacialKeypointDetection(datasets.GeneratorBasedBuilder):
|
|
| 103 |
license=_LICENSE)
|
| 104 |
|
| 105 |
def _split_generators(self, dl_manager):
|
| 106 |
-
# images = dl_manager.download_and_extract(f"{_DATA}images.zip")
|
| 107 |
-
# masks = dl_manager.download_and_extract(f"{_DATA}masks.zip")
|
| 108 |
images = dl_manager.download(f"{_DATA}images.tar.gz")
|
| 109 |
masks = dl_manager.download(f"{_DATA}masks.tar.gz")
|
| 110 |
annotations = dl_manager.download(f"{_DATA}{_NAME}.csv")
|
| 111 |
-
# images = dl_manager.iter_files(images)
|
| 112 |
-
# masks = dl_manager.iter_files(masks)
|
| 113 |
images = dl_manager.iter_archive(images)
|
| 114 |
masks = dl_manager.iter_archive(masks)
|
| 115 |
|
|
@@ -138,30 +71,3 @@ class FacialKeypointDetection(datasets.GeneratorBasedBuilder):
|
|
| 138 |
},
|
| 139 |
'key_points': annotations_df['key_points'].iloc[idx]
|
| 140 |
}
|
| 141 |
-
# images_data = pd.DataFrame(
|
| 142 |
-
# columns=['image_name', 'image_path', 'mask_path'])
|
| 143 |
-
# for idx, ((image_path, image),
|
| 144 |
-
# (mask_path, mask)) in enumerate(zip(images, masks)):
|
| 145 |
-
# images_data.loc[idx] = {
|
| 146 |
-
# 'image_name': image_path.split('/')[-1],
|
| 147 |
-
# 'image_path': image_path,
|
| 148 |
-
# 'mask_path': mask_path
|
| 149 |
-
# }
|
| 150 |
-
|
| 151 |
-
# annotations_df = pd.merge(annotations_df,
|
| 152 |
-
# images_data,
|
| 153 |
-
# how='left',
|
| 154 |
-
# on=['image_name'])
|
| 155 |
-
|
| 156 |
-
# annotations_df[['image_path', 'mask_path'
|
| 157 |
-
# ]] = annotations_df[['image_path',
|
| 158 |
-
# 'mask_path']].astype('string')
|
| 159 |
-
|
| 160 |
-
# for row in annotations_df.sort_values(['image_name'
|
| 161 |
-
# ]).itertuples(index=False):
|
| 162 |
-
# yield idx, {
|
| 163 |
-
# 'image_id': row[0],
|
| 164 |
-
# 'image': row[3],
|
| 165 |
-
# 'mask': row[4],
|
| 166 |
-
# 'key_points': row[2]
|
| 167 |
-
# }
|
|
|
|
| 1 |
import datasets
|
|
|
|
| 2 |
import pandas as pd
|
|
|
|
|
|
|
| 3 |
|
| 4 |
_CITATION = """\
|
| 5 |
@InProceedings{huggingface:dataset,
|
|
|
|
| 24 |
_DATA = f"https://huggingface.co/datasets/TrainingDataPro/{_NAME}/resolve/main/data/"
|
| 25 |
|
| 26 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
class FacialKeypointDetection(datasets.GeneratorBasedBuilder):
|
| 28 |
|
| 29 |
def _info(self):
|
|
|
|
| 40 |
license=_LICENSE)
|
| 41 |
|
| 42 |
def _split_generators(self, dl_manager):
|
|
|
|
|
|
|
| 43 |
images = dl_manager.download(f"{_DATA}images.tar.gz")
|
| 44 |
masks = dl_manager.download(f"{_DATA}masks.tar.gz")
|
| 45 |
annotations = dl_manager.download(f"{_DATA}{_NAME}.csv")
|
|
|
|
|
|
|
| 46 |
images = dl_manager.iter_archive(images)
|
| 47 |
masks = dl_manager.iter_archive(masks)
|
| 48 |
|
|
|
|
| 71 |
},
|
| 72 |
'key_points': annotations_df['key_points'].iloc[idx]
|
| 73 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|