Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Improve oxford_iiit_pet dataset #5349

Open
wants to merge 2 commits into
base: master
Choose a base branch
from
Open

Improve oxford_iiit_pet dataset #5349

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
1ff3092
Improve oxford_iiit_pet dataset:
enricoschroeder Apr 4, 2024
0e9172b
Addressed PR suggestions:
enricoschroeder Apr 26, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
4 changes: 4 additions & 0 deletions tensorflow_datasets/datasets/oxford_iiit_pet/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,7 @@
The Oxford-IIIT pet dataset is a 37 category pet image dataset with roughly 200
images for each class. The images have large variations in scale, pose and
lighting. All images have an associated ground truth annotation of breed.

Please take note of the following:
- Valid bounding boxes of the pets' heads are only provided for the training split. The test split does not contain annotations.
- A few image files in the original dataset are corrupt or in a wrong format. These are being skipped.
1 change: 1 addition & 0 deletions tensorflow_datasets/datasets/oxford_iiit_pet/dummy_data/annotations/xmls/image0_0.xml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
<annotation><folder>OXIIIT</folder><filename>Abyssinian_1.jpg</filename><source><database>OXFORD-IIIT Pet Dataset</database><annotation>OXIIIT</annotation><image>flickr</image></source><size><width>600</width><height>400</height><depth>3</depth></size><segmented>0</segmented><object><name>cat</name><pose>Frontal</pose><truncated>0</truncated><occluded>0</occluded><bndbox><xmin>333</xmin><ymin>72</ymin><xmax>425</xmax><ymax>158</ymax></bndbox><difficult>0</difficult></object></annotation>
1 change: 1 addition & 0 deletions tensorflow_datasets/datasets/oxford_iiit_pet/dummy_data/annotations/xmls/image0_1.xml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
<annotation><folder>OXIIIT</folder><filename>Abyssinian_10.jpg</filename><source><database>OXFORD-IIIT Pet Dataset</database><annotation>OXIIIT</annotation><image>flickr</image></source><size><width>375</width><height>500</height><depth>3</depth></size><segmented>0</segmented><object><name>cat</name><pose>Frontal</pose><truncated>0</truncated><occluded>0</occluded><bndbox><xmin>72</xmin><ymin>105</ymin><xmax>288</xmax><ymax>291</ymax></bndbox><difficult>0</difficult></object></annotation>
1 change: 1 addition & 0 deletions tensorflow_datasets/datasets/oxford_iiit_pet/dummy_data/annotations/xmls/image1_0.xml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
<annotation><folder>OXIIIT</folder><filename>Abyssinian_11.jpg</filename><source><database>OXFORD-IIIT Pet Dataset</database><annotation>OXIIIT</annotation><image>flickr</image></source><size><width>192</width><height>288</height><depth>3</depth></size><segmented>0</segmented><object><name>cat</name><pose>Frontal</pose><truncated>0</truncated><occluded>0</occluded><bndbox><xmin>43</xmin><ymin>57</ymin><xmax>94</xmax><ymax>105</ymax></bndbox><difficult>0</difficult></object></annotation>
1 change: 1 addition & 0 deletions tensorflow_datasets/datasets/oxford_iiit_pet/dummy_data/annotations/xmls/image1_1.xml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
<annotation><folder>OXIIIT</folder><filename>Abyssinian_18.jpg</filename><source><database>OXFORD-IIIT Pet Dataset</database><annotation>OXIIIT</annotation><image>flickr</image></source><size><width>335</width><height>500</height><depth>3</depth></size><segmented>0</segmented><object><name>cat</name><pose>Frontal</pose><truncated>0</truncated><occluded>0</occluded><bndbox><xmin>135</xmin><ymin>87</ymin><xmax>314</xmax><ymax>285</ymax></bndbox><difficult>0</difficult></object></annotation>
1 change: 1 addition & 0 deletions tensorflow_datasets/datasets/oxford_iiit_pet/dummy_data/annotations/xmls/image2_0.xml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
<annotation><folder>OXIIIT</folder><filename>Abyssinian_17.jpg</filename><source><database>OXFORD-IIIT Pet Dataset</database><annotation>OXIIIT</annotation><image>flickr</image></source><size><width>275</width><height>411</height><depth>3</depth></size><segmented>0</segmented><object><name>cat</name><pose>Frontal</pose><truncated>0</truncated><occluded>0</occluded><bndbox><xmin>38</xmin><ymin>92</ymin><xmax>136</xmax><ymax>192</ymax></bndbox><difficult>0</difficult></object></annotation>
1 change: 1 addition & 0 deletions tensorflow_datasets/datasets/oxford_iiit_pet/dummy_data/annotations/xmls/image2_1.xml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
<annotation><folder>OXIIIT</folder><filename>Abyssinian_16.jpg</filename><source><database>OXFORD-IIIT Pet Dataset</database><annotation>OXIIIT</annotation><image>flickr</image></source><size><width>500</width><height>335</height><depth>3</depth></size><segmented>0</segmented><object><name>cat</name><pose>Frontal</pose><truncated>0</truncated><occluded>0</occluded><bndbox><xmin>375</xmin><ymin>53</ymin><xmax>458</xmax><ymax>134</ymax></bndbox><difficult>0</difficult></object></annotation>
1 change: 1 addition & 0 deletions tensorflow_datasets/datasets/oxford_iiit_pet/dummy_data/annotations/xmls/image3_0.xml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
<annotation><folder>OXIIIT</folder><filename>Abyssinian_14.jpg</filename><source><database>OXFORD-IIIT Pet Dataset</database><annotation>OXIIIT</annotation><image>flickr</image></source><size><width>335</width><height>500</height><depth>3</depth></size><segmented>0</segmented><object><name>cat</name><pose>Frontal</pose><truncated>0</truncated><occluded>0</occluded><bndbox><xmin>164</xmin><ymin>71</ymin><xmax>252</xmax><ymax>163</ymax></bndbox><difficult>0</difficult></object></annotation>
1 change: 1 addition & 0 deletions tensorflow_datasets/datasets/oxford_iiit_pet/dummy_data/annotations/xmls/image3_1.xml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
<annotation><folder>OXIIIT</folder><filename>Abyssinian_15.jpg</filename><source><database>OXFORD-IIIT Pet Dataset</database><annotation>OXIIIT</annotation><image>flickr</image></source><size><width>477</width><height>422</height><depth>3</depth></size><segmented>0</segmented><object><name>cat</name><pose>Frontal</pose><truncated>0</truncated><occluded>0</occluded><bndbox><xmin>16</xmin><ymin>16</ymin><xmax>431</xmax><ymax>357</ymax></bndbox><difficult>0</difficult></object></annotation>
1 change: 1 addition & 0 deletions tensorflow_datasets/datasets/oxford_iiit_pet/dummy_data/annotations/xmls/image4_0.xml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
<annotation><folder>OXIIIT</folder><filename>Abyssinian_13.jpg</filename><source><database>OXFORD-IIIT Pet Dataset</database><annotation>OXIIIT</annotation><image>flickr</image></source><size><width>245</width><height>280</height><depth>3</depth></size><segmented>0</segmented><object><name>cat</name><pose>Frontal</pose><truncated>0</truncated><occluded>0</occluded><bndbox><xmin>145</xmin><ymin>25</ymin><xmax>218</xmax><ymax>96</ymax></bndbox><difficult>0</difficult></object></annotation>
1 change: 1 addition & 0 deletions tensorflow_datasets/datasets/oxford_iiit_pet/dummy_data/annotations/xmls/image4_1.xml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
<annotation><folder>OXIIIT</folder><filename>Abyssinian_12.jpg</filename><source><database>OXFORD-IIIT Pet Dataset</database><annotation>OXIIIT</annotation><image>flickr</image></source><size><width>335</width><height>500</height><depth>3</depth></size><segmented>0</segmented><object><name>cat</name><pose>Frontal</pose><truncated>0</truncated><occluded>0</occluded><bndbox><xmin>94</xmin><ymin>83</ymin><xmax>211</xmax><ymax>190</ymax></bndbox><difficult>0</difficult></object></annotation>
55 changes: 54 additions & 1 deletion tensorflow_datasets/datasets/oxford_iiit_pet/oxford_iiit_pet_dataset_builder.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,11 +16,12 @@
"""Oxford-IIIT pet dataset."""

import os
import xml.etree.ElementTree

from tensorflow_datasets.core.utils.lazy_imports_utils import tensorflow as tf
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think tensorflow is only used to open files, could you please replace this with from etils import epath?

import tensorflow_datasets.public_api as tfds

_BASE_URL = "http://www.robots.ox.ac.uk/~vgg/data/pets/data"
_BASE_URL = "https://thor.robots.ox.ac.uk/~vgg/data/pets"

_LABEL_CLASSES = [
"Abyssinian",
Expand Down Expand Up @@ -63,6 +64,44 @@
]
_SPECIES_CLASSES = ["Cat", "Dog"]

# List of samples with corrupt image files
_SKIP_SAMPLES = [
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do you know what the problem with these images is? Could it be that they're encoded in a different format? I suggest we check them in code and try our best to recover, similar to:
https://github.com/SirLefti/datasets/blob/9ae21497aee7177d931e359ffe939e18c7fc1531/tensorflow_datasets/image_classification/cats_vs_dogs.py#L102-L107

"beagle_116",
"chihuahua_121",
"Abyssinian_5",
"Abyssinian_34",
"Egyptian_Mau_14",
"Egyptian_Mau_139",
"Egyptian_Mau_145",
"Egyptian_Mau_156",
"Egyptian_Mau_167",
"Egyptian_Mau_177",
"Egyptian_Mau_186",
"Egyptian_Mau_191"
]

def _get_head_bbox(annon_filepath):
"""Read head bbox from annotation XML file."""
with tf.io.gfile.GFile(annon_filepath, "r") as f:
root = xml.etree.ElementTree.parse(f).getroot()

# Disable pytype to avoid attribute-error due to find returning
# Optional[Element]
# pytype: disable=attribute-error
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This disables pytype for the rest of the file, which is undesirable. How about something like size: xml.etree.ElementTree.Element = root.find("size") # pytype: disable=annotation-type-mismatch?

size = root.find("size")
width = float(size.find("width").text)
height = float(size.find("height").text)

obj = root.find("object")
bndbox = obj.find("bndbox")
xmax = float(bndbox.find("xmax").text)
xmin = float(bndbox.find("xmin").text)
ymax = float(bndbox.find("ymax").text)
ymin = float(bndbox.find("ymin").text)
return tfds.features.BBox(
ymin / height, xmin / width, ymax / height, xmax / width
)


class Builder(tfds.core.GeneratorBasedBuilder):
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please bump VERSION and add RELEASE_NOTES.

"""Oxford-IIIT pet dataset."""
Expand All @@ -79,6 +118,7 @@ def _info(self):
"segmentation_mask": tfds.features.Image(
shape=(None, None, 1), use_colormap=True
),
"head": tfds.features.BBoxFeature()
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

How about a more explicit feature name, e.g. "head_bbox"?

}),
supervised_keys=("image", "label"),
homepage="http://www.robots.ox.ac.uk/~vgg/data/pets/",
Expand Down Expand Up @@ -126,18 +166,31 @@ def _generate_examples(
for line in images_list:
image_name, label, species, _ = line.strip().split(" ")

# skip corrupt samples
if image_name in _SKIP_SAMPLES:
continue

trimaps_dir_path = os.path.join(annotations_dir_path, "trimaps")
xmls_dir_path = os.path.join(annotations_dir_path, "xmls")

trimap_name = image_name + ".png"
xml_name = image_name + ".xml"
image_name += ".jpg"
label = int(label) - 1
species = int(species) - 1

try:
head_bbox = _get_head_bbox(os.path.join(xmls_dir_path, xml_name))
except tf.errors.NotFoundError:
# test samples do not have an annotation file
head_bbox = tfds.features.BBox(0., 0., 0., 0.)
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Move this to a constant _EMPTY_BBOX.


record = {
"image": os.path.join(images_dir_path, image_name),
"label": int(label),
"species": species,
"file_name": image_name,
"segmentation_mask": os.path.join(trimaps_dir_path, trimap_name),
"head": head_bbox
}
yield image_name, record