def test_catch_invalid_scene_name():
"""Test catching invalid scene name"""
scene_name = "Invalid scene name"
with pytest.raises(InvalidSceneName):
InvalidSceneName.validate_scene_list(scene_name,
Dataset.get_scene_list())
with pytest.raises(InvalidSceneName):
InvalidSceneName.validate_scene_info_list(
scene_name, Dataset.get_training_scene_list())
def test_2003_disparity():
"""Test 2003 load disparity image"""
ssl._create_default_https_context = ssl._create_unverified_context
DATASET_FOLDER = os.path.join(os.getcwd(),"datasets") #Path to download datasets
if not os.path.exists(DATASET_FOLDER):
os.makedirs(DATASET_FOLDER)
scene_info = SceneInfo(Dataset.Teddy, DatasetType.imperfect, 1.0)
scene_name = scene_info.scene_name
dataset_type = scene_info.dataset_type
Dataset.download_scene_data(scene_name,DATASET_FOLDER,dataset_type)
Dataset.load_scene_data(
scene_name=scene_name,dataset_folder=DATASET_FOLDER,
dataset_type=dataset_type)
def test_catch_invalid_scene_name():
"""Test valid scene names"""
with pytest.raises(InvalidSceneName):
InvalidSceneName.validate_scene_list("INVALID SCENE NAME", Dataset.get_scene_list())
def test_check_valid_scene_names():
"""Test valid scene names"""
for scene_name in Dataset.get_scene_list():
assert InvalidSceneName.validate_scene_list(scene_name, Dataset.get_scene_list()) is None
def test_dataset_valid_scene_urls():
"""Test valid url creation for scenes"""
for scene_name in Dataset.get_scene_list():
url = Dataset.get_url_from_scene(scene_name)
assert validators.url(url)
def test_init_dataset():
"""Test initalising Dataset class"""
Dataset()
This module demonstrates loading data from one scene using the stereomideval module.
"""
import os
from stereomideval.dataset import Dataset
# Path to dowmload datasets
DATASET_FOLDER = os.path.join(os.getcwd(), "datasets")
# Scene name (see here for list of scenes: https://vision.middlebury.edu/stereo/data/scenes2014/)
SCENE_NAME = "Adirondack"
# Display loaded scene to OpenCV window
DISPLAY_IMAGES = False
# Create dataset folder
if not os.path.exists(DATASET_FOLDER):
os.makedirs(DATASET_FOLDER)
# Download dataset from middlebury servers
# will only download it if it hasn't already been downloaded
print("Downloading data for scene '" + SCENE_NAME + "'...")
Dataset.download_scene_data(SCENE_NAME, DATASET_FOLDER)
# Load scene data from downloaded folder
print("Loading data for scene '" + SCENE_NAME + "'...")
scene_data = Dataset.load_scene_data(SCENE_NAME, DATASET_FOLDER,
DISPLAY_IMAGES)
# Scene data class contains the following data:
left_image = scene_data.left_image
right_image = scene_data.right_image
ground_truth_disp_image = scene_data.disp_image
ndisp = scene_data.ndisp
This module demonstrates loading data from all scenes using the stereomideval module.
"""
import os
from stereomideval.dataset import Dataset
DATASET_FOLDER = os.path.join(os.getcwd(),
"datasets") #Path to download datasets
DISPLAY_IMAGES = False
# Create dataset folder
if not os.path.exists(DATASET_FOLDER):
os.makedirs(DATASET_FOLDER)
# Get list of scenes in Milddlebury's stereo training dataset and iterate through them
for scene_info in Dataset.get_training_scene_list():
scene_name = scene_info.scene_name
dataset_type = scene_info.dataset_type
# Download dataset from middlebury servers
# will only download it if it hasn't already been downloaded
print("Downloading data for scene '" + scene_name + "'...")
Dataset.download_scene_data(scene_name, DATASET_FOLDER, dataset_type)
# Load scene data from downloaded folder
print("Loading data for scene '" + scene_name + "'...")
scene_data = Dataset.load_scene_data(scene_name=scene_name,
dataset_folder=DATASET_FOLDER,
dataset_type=dataset_type,
display_images=DISPLAY_IMAGES)
# Scene data class contains the following data:
left_image = scene_data.left_image
right_image = scene_data.right_image
from stereomideval.structures import MatchData
from stereomideval.dataset import Dataset
from stereomideval.eval import Eval, Timer
DATASET_FOLDER = os.path.join(os.getcwd(),
"datasets") #Path to download datasets
GET_METRIC_RANK = False # Compare each match data against online ranking
GET_AV_METRIC_RANK = True # Compare average results across all scenes against online ranking
# Create dataset folder
if not os.path.exists(DATASET_FOLDER):
os.makedirs(DATASET_FOLDER)
match_data_list = []
# Get list of scenes in Milddlebury's stereo training dataset and iterate through them
for scene_info in Dataset.get_training_scene_list():
scene_name = scene_info.scene_name
dataset_type = scene_info.dataset_type
# Download dataset from middlebury servers
# will only download it if it hasn't already been downloaded
print("Downloading data for scene '" + scene_name + "'...")
Dataset.download_scene_data(scene_name, DATASET_FOLDER, dataset_type)
# Load scene data from downloaded folder
print("Loading data for scene '" + scene_name + "'...")
scene_data = Dataset.load_scene_data(scene_name=scene_name,
dataset_folder=DATASET_FOLDER,
dataset_type=dataset_type)
# Scene data class contains the following data:
left_image = scene_data.left_image
right_image = scene_data.right_image
ground_truth_disp_image = scene_data.disp_image
def get_metric_vals(metric,
scene_name,
dataset_type=DatasetType.imperfect,
dense=True):
"""
Get list of metrics values from Middlebury website
Currently assumes using training dense set with nonocc mask
Parameters:
metric (Eval.Metric): metric list to get values for
scene_name (string): scene name to get values for
Returns:
alg_names (list(string)): list of algorithm names
metric_list (list(string)): list of values for given metric
"""
# Check scene name is valid
InvalidSceneName.validate_scene_info_list(
scene_name, Dataset.get_training_scene_list())
dense_or_sparse = "dense"
if not dense:
dense_or_sparse = "sparse"
# Get url for table on Middlebury website
url = WebscrapeMiddlebury.get_table_url("training", dense_or_sparse,
metric, "nonocc", False)
# disable warnings to avoid message:
# 'InsecureRequestWarning: Unverified HTTPS request
# is being made to host 'vision.middlebury.edu'.
# Adding certificate verification is strongly advised.'
urllib3.disable_warnings()
# Load url
print("Loading results from Middlebury website...")
page = requests.get(url, verify=False)
# Parse webpage
soup = BeautifulSoup(page.content, 'html.parser')
# Get stereo table from webpage
table = soup.find('table', {'id': 'stereoTable'})
# Find table body
table_body = table.find('tbody')
# Find table body rows
table_rows = table_body.findAll('tr')
# Initalise lists for storing result of table
alg_names_list = []
metric_list = []
web_scene_name_suffix = WebscrapeMiddlebury.get_web_scene_name_suffix(
dataset_type)
# (e.g. trying to compare image that is not on website table)
# Iteration through rows in table
for table_row in list(table_rows):
# Find algorithm name in row data
alg_name = table_row.find('td', {'class': ['algName']})
# Find metric value in row data
metric_val = table_row.find('td', {
"class": [
"{} data datasetCol".format(scene_name +
web_scene_name_suffix)
]
},
partial=False)
# Check if metric value was found
if metric_val is None:
# Find metric value with 'firstPlace'
# (the first row in the data has this extra class definition)
metric_val = table_row.find('td', {
"class": [
"{} firstPlace data datasetCol".format(
scene_name + web_scene_name_suffix)
]
},
partial=False)
# Check metric value and algorithm name were found
if metric_val is not None and alg_name is not None:
# Add metric value and algorithm name to lists
metric_list.append(float(metric_val.string))
alg_names_list.append(alg_name.string)
else:
raise Exception(
"Failed to find table data for metric value or algorithm name"
)
return alg_names_list, metric_list