def configure_experiments():
""" This function is automatically called by
the command ./manage.py mtconfigure """
# must be imported locally to avoid a circular import
from mturk.utils import configure_experiment
# aliases
sandbox = settings.MTURK_SANDBOX
production = not sandbox
configure_experiment(
slug='quality_scene',
template_dir='photos/experiments',
module='photos.experiments',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.02'),
num_outputs_max=5,
contents_per_hit=50,
content_type_model=Photo,
out_content_type_model=PhotoSceneQualityLabel,
out_content_attr='photo',
content_filter='{}',
title='Click on images that belong to a scene category',
description='Given a collection of images, your job is to decide which images '
'belong to that scene category.',
keywords='category,images,collection,image,picture,classify',
#frame_height=8000,
requirements={},
auto_add_hits=False, # settings.MTURK_SANDBOX,
)
configure_experiment(
slug='label_whitebalance',
template_dir='photos/experiments',
module='photos.experiments',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.05'),
num_outputs_max=5,
contents_per_hit=15 if production else 4,
max_active_hits=50,
content_type_model=Photo,
out_content_type_model=PhotoWhitebalanceLabel,
out_content_attr='photo',
content_filter={
'scene_category_correct': True,
'num_shapes__gt': 0,
},
title='Click on white or gray things in images',
description='Your task is to click on points in an image that are white or gray.',
keywords='whitebalance,appearance,image,picture,label',
requirements={
'min_points': 3
},
auto_add_hits=False, # settings.MTURK_SANDBOX,
)
def configure_experiments():
""" This function is automatically called by
the command ./manage.py mtconfigure """
# must be imported locally to avoid a circular import
from mturk.utils import configure_experiment
# aliases
sandbox = settings.MTURK_SANDBOX
production = not sandbox
configure_experiment(
slug='quality_scene',
template_dir='photos/experiments',
module='photos.experiments',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.02'),
num_outputs_max=5,
contents_per_hit=50,
content_type_model=Photo,
out_content_type_model=PhotoSceneQualityLabel,
out_content_attr='photo',
content_filter='{}',
title='Click on images that belong to a scene category',
description=
'Given a collection of images, your job is to decide which images '
'belong to that scene category.',
keywords='category,images,collection,image,picture,classify',
#frame_height=8000,
requirements={},
auto_add_hits=False, # settings.MTURK_SANDBOX,
)
configure_experiment(
slug='label_whitebalance',
template_dir='photos/experiments',
module='photos.experiments',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.05'),
num_outputs_max=5,
contents_per_hit=15 if production else 4,
max_active_hits=50,
content_type_model=Photo,
out_content_type_model=PhotoWhitebalanceLabel,
out_content_attr='photo',
content_filter={
'scene_category_correct': True,
'num_shapes__gt': 0,
},
title='Click on white or gray things in images',
description=
'Your task is to click on points in an image that are white or gray.',
keywords='whitebalance,appearance,image,picture,label',
requirements={'min_points': 3},
auto_add_hits=False, # settings.MTURK_SANDBOX,
)
def configure_experiments():
""" This function is automatically called by
the command ./manage.py mtconfigure """
# must be imported locally to avoid a circular import
from mturk.utils import configure_experiment
# aliases
sandbox = settings.MTURK_SANDBOX
production = not sandbox
configure_experiment(
slug='segment_material',
template_dir='shapes/experiments',
module='shapes.experiments',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.12'),
num_outputs_max=6,
out_count_ratio=6,
contents_per_hit=1,
max_active_hits=200,
content_type_model=Photo,
out_content_type_model=SubmittedShape,
out_content_attr='photo',
content_filter={
'scene_category_correct': True,
'license__creative_commons': True,
'license__cc_no_deriv': False,
},
duration=60 * 60 * 3,
title='Draw shapes in images',
description=
'Draw polygons around regions that have the same material in a photograph.',
keywords='material,substance,polygon,shape,draw,image,picture,classify',
requirements={
'min_shapes': 6,
'min_vertices': 3
},
qualifications='{ "mat_seg": 1 }',
auto_add_hits=False, # settings.MTURK_SANDBOX,
)
configure_experiment(
slug='quality_material',
template_dir='shapes/experiments',
module='shapes.experiments',
examples_group_attr='shape',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.04'),
num_outputs_max=5 if production else 10,
contents_per_hit=40,
content_type_model=MaterialShape,
out_content_attr='shape',
out_content_type_model=MaterialShapeQuality,
content_filter={
'invalid': False,
'correct__isnull': True,
'pixel_area__gt': Shape.MIN_PIXEL_AREA,
'photo__whitebalanced': True,
'photo__scene_category_correct': True,
},
title='Click on images that contain a single material',
description=
'This task involves clicking on images that contain a single type of material or texture within a red boundary.',
keywords='material,substance,shape,image,picture,classify,label',
requirements={},
auto_add_hits=False, # settings.MTURK_SANDBOX,
#examples_good=MaterialShape.objects.filter(
#qualities__user=examples_user,
#qualities__correct=True).order_by('?')[:200],
#examples_bad=MaterialShape.objects.filter(
#qualities__user=examples_user,
#qualities__correct=False).order_by('?')[:200],
)
configure_experiment(
slug='label_substance',
template_dir='shapes/experiments',
module='shapes.experiments',
examples_group_attr='shape',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.10'),
num_outputs_max=5,
min_output_consensus=3,
contents_per_hit=40,
#max_active_hits=200,
content_type_model=MaterialShape,
out_content_type_model=ShapeSubstanceLabel,
out_content_attr='shape',
content_filter={
'invalid': False,
'correct': True,
'pixel_area__gt': Shape.MIN_PIXEL_AREA,
'num_vertices__gte': 4,
#'photo__whitebalanced': True,
'photo__scene_category_correct': True,
},
title='Click on the name of a material in an image',
description=
'Given a shape highlighted in an image, select its material from a list.',
keywords='material,substance,appearance,image,picture,classify',
frame_height=1200,
requirements={},
#qualifications='{ "substance": 1 }',
auto_add_hits=False, # settings.MTURK_SANDBOX,
)
configure_experiment(
slug='label_planarity',
template_dir='shapes/experiments',
module='shapes.experiments',
examples_group_attr='shape',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.04'),
num_outputs_max=5 if production else 10,
contents_per_hit=40,
content_type_model=MaterialShape,
out_content_type_model=ShapePlanarityLabel,
out_content_attr='shape',
content_filter={
'invalid': False,
'pixel_area__gt': Shape.MIN_PLANAR_AREA,
'correct': True,
'substance__isnull': False,
'substance__fail': False,
'num_vertices__gte': 4,
#'photo__whitebalanced': True,
'photo__scene_category_correct': True,
},
title='Click on images of planar objects',
description=
'Your task is to select the planar objects out of a collection of images.',
keywords=
'flat,planar,appearance,polygon,shape,draw,image,picture,classify,label',
#frame_height=7500,
requirements={},
auto_add_hits=False, # settings.MTURK_SANDBOX,
#examples_good=MaterialShape.objects.filter(
#planarities__user=examples_user,
#planarities__planar=True).order_by('?')[:200],
#examples_bad=MaterialShape.objects.filter(
#planarities__user=examples_user,
#planarities__planar=False).order_by('?')[:200],
)
for group in ShapeSubstanceGroup.objects.filter(active=True):
configure_experiment(
slug='label_name',
template_dir='shapes/experiments',
module='shapes.experiments',
examples_group_attr='shape',
variant=json.dumps({'substance_group_id': group.id}),
completed_id='label_name',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.10'),
num_outputs_max=5,
min_output_consensus=3,
contents_per_hit=50,
content_type_model=MaterialShape,
out_content_type_model=MaterialShapeNameLabel,
out_content_attr='shape',
content_filter={
'invalid': False,
'pixel_area__gt': Shape.MIN_PIXEL_AREA,
'correct': True,
'substance__isnull': False,
'substance__group_id': group.id,
'photo__scene_category_correct': True,
'num_vertices__gte': 4,
#'photo__scene_category__name': 'kitchen',
},
frame_height=1200,
title='Click on the name of an object in an image',
description=
'Given an object highlighted in an image, select its common English name from a list. (id: %s)'
% group.id,
keywords='object,categorize,appearance,image,picture,classify',
requirements={},
#qualifications='{ "substance": 1 }',
auto_add_hits=
False, # settings.MTURK_SANDBOX, # TODO: get rid of fixed list
)
def configure_experiments():
""" This function is automatically called by
the command ./manage.py mtconfigure """
from mturk.utils import configure_experiment
production = not settings.MTURK_SANDBOX
configure_experiment(
slug='intrinsic_opacity',
template_dir='intrinsic/experiments',
module='intrinsic.experiments',
version=2, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.05'),
num_outputs_max=5,
contents_per_hit=200, # if production else 50,
frame_height=668, # handle 1024x768 window sizes
content_type_model=IntrinsicPoint,
out_content_type_model=IntrinsicPointOpacityResponse,
out_content_attr='point',
content_filter={
#'photo__license__publishable': True,
#'photo__stylized': False,
#'photo__rotated': False,
#'photo__nonperspective': False,
#'photo__synthetic': True,
'photo__intrinsic_synthetic__multilayer_exr__isnull': False,
#'photo__scene_category_correct': True,
#'photo__num_intrinsic_points__gte': 20,
#'min_separation__lt': 0.05,
},
title='Click on mirrors and transparent surfaces',
description=
'Given points in images, your job is to decide which belong to mirrors or transparent objects.',
keywords='category,images,collection,image,picture,classify,color',
requirements={},
auto_add_hits=True,
has_tutorial=True,
)
# don't add photo 10386; it's somewhat ambiguous
# don't add photo 94610 -- it's all gloss and metal
excluded_photo_ids = [10386, 94610]
# hard-code examples here for now
equal_tests = IntrinsicPointComparison.objects.in_bulk(
list(
set([
961869,
961802,
961774,
961815,
961848,
961789,
961814,
961888,
961868,
961885,
961713,
961742,
961818,
961813,
961864,
961807,
961743,
961766,
961784,
961884,
961714,
999517,
999510,
1006818,
1006825,
1006838,
1006808,
1006820,
1006810,
1046422,
1046444,
1046369,
1046437,
1046352,
1046338,
1046425,
1046459,
929022,
929035,
929037,
929031,
929008,
928985,
929006,
929035,
2385973,
2386026,
2385959,
2386106,
2385974,
1605947,
1605914,
1605960,
1605989,
1605962,
1606032,
1605956,
1605931,
1605946,
1605988,
1606043,
1606044,
1118334,
1118351,
1118282,
1118347,
1118422,
1118329,
1118260,
1118410,
1118350,
1118443,
1118311,
1118306,
1118404,
1118362,
1118286,
1118405,
1118345,
1118287,
1118335,
1118367,
1118395,
1118390,
1118308,
1118400,
1118434,
1118301,
1118425,
1118439,
1118369,
1118415,
1118281,
1118288,
1118362,
2419322,
2419219,
2419210,
2419204,
956745,
956741,
956641,
928997,
928997,
928981,
1720063,
1720071,
1719956,
1719978,
1720041,
1720042,
1719928,
1720011,
1719905,
1719973,
1720059,
1719918,
1720012,
1720080,
1719915,
1720025,
1719974,
1719943,
1719908,
1719922,
1720044,
1720013,
1402497,
1402598,
1402545,
1719916,
1719958,
1720078,
1720628,
1720590,
1720606,
1720641,
1320564,
1320545,
1320547,
1320548,
2432036,
3708799,
3445549,
1118283,
1118258,
2508708,
951758,
951741,
951662,
951758,
951717,
951620,
951712,
951629,
]))).values()
for comparison in equal_tests:
if comparison.photo_id not in excluded_photo_ids:
comparison.darker = 'E'
comparison.darker_method = 'A'
comparison.save()
point1_tests = IntrinsicPointComparison.objects.in_bulk([
961811, 961865, 961759, 999516, 999490, 999491, 999583, 999500, 999564,
1046371, 2358770, 2358731, 1118331, 1118290, 1118263, 1118302, 2358729,
2358731, 2358695, 1180756, 1180681, 1180720, 1180701, 1180673, 976878,
976878, 999565, 1720053, 1719988, 1719951, 1402559, 1720077, 1720629,
1320551, 1320549, 1320627, 1320606, 1320622, 2431927, 2432017, 2431952,
2432053, 2431947, 2432055, 3708827, 3708860, 3708802, 3708808, 3708899,
3708874, 3708845, 3445595, 3445598, 3445566, 3445584, 3445498, 3708895,
3708908, 3708894, 3889277, 3889277, 3889172, 961827, 961769, 961891,
961725, 961744, 2369862, 2369839, 2369873, 2508687, 2508663, 2508695,
2508631
]).values()
for comparison in point1_tests:
if comparison.photo_id not in excluded_photo_ids:
comparison.darker = '1'
comparison.darker_method = 'A'
comparison.save()
point2_tests = IntrinsicPointComparison.objects.in_bulk([
999515, 999498, 999499, 999569, 999562, 1046453, 2358755, 2358801,
2358691, 1118365, 1118418, 2358714, 2358755, 956602, 956601, 1180669,
1180730, 976827, 976825, 1719954, 1720019, 1402521, 1402562, 1720565,
1320579, 2431925, 2431989, 2431923, 2432011, 3708773, 3708878, 3445472,
3445554, 3445586, 3445550, 3445609, 3445556, 3445546, 3889199, 3889278,
3889145, 2208252, 2369764, 2508705, 2508711
]).values()
for comparison in point2_tests:
if comparison.photo_id not in excluded_photo_ids:
comparison.darker = '2'
comparison.darker_method = 'A'
comparison.save()
test_contents = [
t for t in (equal_tests + point1_tests + point2_tests)
if t.photo_id not in excluded_photo_ids
]
configure_experiment(
slug='intrinsic_compare',
template_dir='intrinsic/experiments',
module='intrinsic.experiments',
version=2, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.06'),
num_outputs_max=5,
contents_per_hit=25, # if production else 2,
max_active_hits=10000 if production else 50,
frame_height=668, # handle 1024x768 window sizes
content_type_model=IntrinsicPointComparison,
out_content_type_model=IntrinsicPointComparisonResponse,
out_content_attr='comparison',
content_filter={
'point1__opaque': True,
'point2__opaque': True,
'photo__intrinsic_synthetic__multilayer_exr__isnull': False,
#'photo__license__publishable': True,
#'photo__stylized': False,
#'photo__rotated': False,
#'photo__nonperspective': False,
#'photo__synthetic': False,
#'photo__scene_category_correct': True,
#'point1__min_separation__lt': 0.05,
},
title='Compare colors in an image',
description=
'Given pairs of points in images, your job is to decide which of two points is darker.',
keywords='category,images,collection,image,picture,classify,color',
requirements={},
auto_add_hits=True,
has_tutorial=True,
test_contents=test_contents,
test_contents_per_assignment=5,
)
def configure_experiments():
""" This function is automatically called by
the command ./manage.py mtconfigure """
# must be imported locally to avoid a circular import
from mturk.utils import configure_experiment
# aliases
sandbox = settings.MTURK_SANDBOX
production = not sandbox
# set up envmaps
envmap = EnvironmentMap.objects.get_or_create(
user=User.objects.get_or_create(
username='******')[0].get_profile(),
name='ennis')
for envmap in EnvironmentMap.objects.all():
configure_experiment(
slug='bsdf_wd',
template_dir='bsdfs/experiments',
module='bsdfs.experiments',
examples_group_attr='shape',
variant={'envmap_id': envmap.id},
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.10'),
num_outputs_max=1,
contents_per_hit=10,
max_active_hits=2000,
content_type_model=MaterialShape,
out_content_type_model=ShapeBsdfLabel_wd,
out_content_attr='shape',
content_filter={
#'synthetic': True,
#'synthetic_slug__in': ['teapot', 'teacup', 'spoon', 'coyote'],
'invalid': False,
'pixel_area__gt': Shape.MIN_PIXEL_AREA,
'num_vertices__gte': 10,
'correct': True,
'substance__isnull': False,
'substance__fail': False,
'photo__whitebalanced': True,
'photo__scene_category_correct': True,
},
title='Adjust a blob to match an image',
description='Looking at an image, your goal is to adjust the appearance '
'of a blob so that it matches a target photograph. A modern '
'browser is required.',
keywords='material,appearance,image,picture,classify,BRDF,microfacet,blob,appearance',
frame_height=1150,
requirements={},
#qualifications='{ "bsdf_match": 1 }',
auto_add_hits=False, # settings.MTURK_SANDBOX,
)
for attr in ('color', 'gloss'):
content_filter = {
'invalid': False,
'shape__invalid': False,
'give_up': False,
#'shape__pixel_area__gt': Shape.MIN_PIXEL_AREA,
#'shape__correct': True,
#'shape__substance__isnull': False,
#'shape__substance__fail': False,
#'shape__photo__whitebalanced': True,
#'shape__photo__scene_category_correct': True,
}
if production and attr == 'gloss':
content_filter['color_correct'] = True
configure_experiment(
slug='quality_bsdf_%s' % attr,
template_dir='bsdfs/experiments',
module='bsdfs.experiments',
examples_group_attr='shape',
variant={'bsdf_version': 'wd'},
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.04'),
num_outputs_max=5,
contents_per_hit=40,
content_type_model=ShapeBsdfLabel_wd,
out_content_type_model=ShapeBsdfQuality,
out_content_attr='shapebsdflabel_wd',
content_filter=content_filter,
title='Click on blobs that match an image (%s)' % attr,
description='This task involves clicking on images that match a blob next to the image.',
keywords='material,substance,shape,image,picture,classify,label,blob,match,appearance',
#frame_height=7500,
requirements={},
auto_add_hits=False, # settings.MTURK_SANDBOX,
)
def configure_experiments():
""" This function is automatically called by
the command ./manage.py mtconfigure """
# must be imported locally to avoid a circular import
from mturk.utils import configure_experiment
# aliases
sandbox = settings.MTURK_SANDBOX
production = not sandbox
configure_experiment(
slug='segment_material',
template_dir='shapes/experiments',
module='shapes.experiments',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.12'),
num_outputs_max=6,
out_count_ratio=6,
contents_per_hit=1,
max_active_hits=200,
content_type_model=Photo,
out_content_type_model=SubmittedShape,
out_content_attr='photo',
content_filter={
'scene_category_correct': True,
'license__creative_commons': True,
'license__cc_no_deriv': False,
},
duration=60 * 60 * 3,
title='Draw shapes in images',
description='Draw polygons around regions that have the same material in a photograph.',
keywords='material,substance,polygon,shape,draw,image,picture,classify',
requirements={
'min_shapes': 6,
'min_vertices': 3
},
qualifications='{ "mat_seg": 1 }',
auto_add_hits=False, # settings.MTURK_SANDBOX,
)
configure_experiment(
slug='quality_material',
template_dir='shapes/experiments',
module='shapes.experiments',
examples_group_attr='shape',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.04'),
num_outputs_max=5 if production else 10,
contents_per_hit=40,
content_type_model=MaterialShape,
out_content_attr='shape',
out_content_type_model=MaterialShapeQuality,
content_filter={
'invalid': False,
'correct__isnull': True,
'pixel_area__gt': Shape.MIN_PIXEL_AREA,
'photo__whitebalanced': True,
'photo__scene_category_correct': True,
},
title='Click on images that contain a single material',
description='This task involves clicking on images that contain a single type of material or texture within a red boundary.',
keywords='material,substance,shape,image,picture,classify,label',
requirements={},
auto_add_hits=False, # settings.MTURK_SANDBOX,
#examples_good=MaterialShape.objects.filter(
#qualities__user=examples_user,
#qualities__correct=True).order_by('?')[:200],
#examples_bad=MaterialShape.objects.filter(
#qualities__user=examples_user,
#qualities__correct=False).order_by('?')[:200],
)
configure_experiment(
slug='label_substance',
template_dir='shapes/experiments',
module='shapes.experiments',
examples_group_attr='shape',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.10'),
num_outputs_max=5,
min_output_consensus=3,
contents_per_hit=40,
#max_active_hits=200,
content_type_model=MaterialShape,
out_content_type_model=ShapeSubstanceLabel,
out_content_attr='shape',
content_filter={
'invalid': False,
'correct': True,
'pixel_area__gt': Shape.MIN_PIXEL_AREA,
'num_vertices__gte': 4,
#'photo__whitebalanced': True,
'photo__scene_category_correct': True,
},
title='Click on the name of a material in an image',
description='Given a shape highlighted in an image, select its material from a list.',
keywords='material,substance,appearance,image,picture,classify',
frame_height=1200,
requirements={},
#qualifications='{ "substance": 1 }',
auto_add_hits=False, # settings.MTURK_SANDBOX,
)
configure_experiment(
slug='label_planarity',
template_dir='shapes/experiments',
module='shapes.experiments',
examples_group_attr='shape',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.04'),
num_outputs_max=5 if production else 10,
contents_per_hit=40,
content_type_model=MaterialShape,
out_content_type_model=ShapePlanarityLabel,
out_content_attr='shape',
content_filter={
'invalid': False,
'pixel_area__gt': Shape.MIN_PLANAR_AREA,
'correct': True,
'substance__isnull': False,
'substance__fail': False,
'num_vertices__gte': 4,
#'photo__whitebalanced': True,
'photo__scene_category_correct': True,
},
title='Click on images of planar objects',
description='Your task is to select the planar objects out of a collection of images.',
keywords='flat,planar,appearance,polygon,shape,draw,image,picture,classify,label',
#frame_height=7500,
requirements={},
auto_add_hits=False, # settings.MTURK_SANDBOX,
#examples_good=MaterialShape.objects.filter(
#planarities__user=examples_user,
#planarities__planar=True).order_by('?')[:200],
#examples_bad=MaterialShape.objects.filter(
#planarities__user=examples_user,
#planarities__planar=False).order_by('?')[:200],
)
for group in ShapeSubstanceGroup.objects.filter(active=True):
configure_experiment(
slug='label_name',
template_dir='shapes/experiments',
module='shapes.experiments',
examples_group_attr='shape',
variant=json.dumps({'substance_group_id': group.id}),
completed_id='label_name',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.10'),
num_outputs_max=5,
min_output_consensus=3,
contents_per_hit=50,
content_type_model=MaterialShape,
out_content_type_model=MaterialShapeNameLabel,
out_content_attr='shape',
content_filter={
'invalid': False,
'pixel_area__gt': Shape.MIN_PIXEL_AREA,
'correct': True,
'substance__isnull': False,
'substance__group_id': group.id,
'photo__scene_category_correct': True,
'num_vertices__gte': 4,
#'photo__scene_category__name': 'kitchen',
},
frame_height=1200,
title='Click on the name of an object in an image',
description='Given an object highlighted in an image, select its common English name from a list. (id: %s)' % group.id,
keywords='object,categorize,appearance,image,picture,classify',
requirements={},
#qualifications='{ "substance": 1 }',
auto_add_hits=False, # settings.MTURK_SANDBOX, # TODO: get rid of fixed list
)
def configure_experiments():
""" This function is automatically called by
the command ./manage.py mtconfigure """
# must be imported locally to avoid a circular import
from mturk.utils import configure_experiment
# aliases
sandbox = settings.MTURK_SANDBOX
production = not sandbox
# full person annotation
configure_experiment(
slug='segment_person',
variant='"person_fix"',
template_dir='segmentation/experiments',
module='segmentation.experiments',
version=2, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.14'),
num_outputs_max=1,
contents_per_hit=2,
content_type_model=PersonSegmentationTask,
out_content_type_model=PersonSegmentation,
out_content_attr='task',
content_filter={
'responses__isnull': True,
'part__isnull': True,
},
title='Carefully segment a person',
description=
'Given an image, your job is to segment a person from an image.',
keywords='person,images,segment',
#frame_height=8000,
requirements={},
auto_add_hits=False,
has_tutorial=True,
)
# part annotation
configure_experiment(
slug='segment_part_person',
variant='',
template_dir='segmentation/experiments',
module='segmentation.experiments',
version=2, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.11'),
num_outputs_max=1,
contents_per_hit=3,
content_type_model=PersonSegmentationTask,
out_content_type_model=PersonSegmentation,
out_content_attr='task',
content_filter={
# only take tasks where we want to segment a part
'part__isnull': False,
},
title='Carefully segment a part of a person',
description='Given an image, your job is to segment a part of a person '
+ 'from an image.',
keywords='person,part,images,segment',
#frame_height=8000,
requirements={},
auto_add_hits=False,
has_tutorial=True,
)
def configure_experiments():
""" This function is automatically called by
the command ./manage.py mtconfigure """
# must be imported locally to avoid a circular import
from mturk.utils import configure_experiment
# aliases
sandbox = settings.MTURK_SANDBOX
production = not sandbox
# set up envmaps
envmap = EnvironmentMap.objects.get_or_create(
user=User.objects.get_or_create(username="******")[0].get_profile(), name="ennis"
)
for envmap in EnvironmentMap.objects.all():
configure_experiment(
slug="bsdf_wd",
template_dir="bsdfs/experiments",
module="bsdfs.experiments",
examples_group_attr="shape",
variant={"envmap_id": envmap.id},
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal("0.10"),
num_outputs_max=1,
contents_per_hit=10,
max_active_hits=2000,
content_type_model=MaterialShape,
out_content_type_model=ShapeBsdfLabel_wd,
out_content_attr="shape",
content_filter={
#'synthetic': True,
#'synthetic_slug__in': ['teapot', 'teacup', 'spoon', 'coyote'],
"invalid": False,
"pixel_area__gt": Shape.MIN_PIXEL_AREA,
"num_vertices__gte": 10,
"correct": True,
"substance__isnull": False,
"substance__fail": False,
"photo__whitebalanced": True,
"photo__scene_category_correct": True,
},
title="Adjust a blob to match an image",
description="Looking at an image, your goal is to adjust the appearance "
"of a blob so that it matches a target photograph. A modern "
"browser is required.",
keywords="material,appearance,image,picture,classify,BRDF,microfacet,blob,appearance",
frame_height=1150,
requirements={},
# qualifications='{ "bsdf_match": 1 }',
auto_add_hits=False, # settings.MTURK_SANDBOX,
)
for attr in ("color", "gloss"):
content_filter = {
"invalid": False,
"shape__invalid": False,
"give_up": False,
#'shape__pixel_area__gt': Shape.MIN_PIXEL_AREA,
#'shape__correct': True,
#'shape__substance__isnull': False,
#'shape__substance__fail': False,
#'shape__photo__whitebalanced': True,
#'shape__photo__scene_category_correct': True,
}
if production and attr == "gloss":
content_filter["color_correct"] = True
configure_experiment(
slug="quality_bsdf_%s" % attr,
template_dir="bsdfs/experiments",
module="bsdfs.experiments",
examples_group_attr="shape",
variant={"bsdf_version": "wd"},
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal("0.04"),
num_outputs_max=5,
contents_per_hit=40,
content_type_model=ShapeBsdfLabel_wd,
out_content_type_model=ShapeBsdfQuality,
out_content_attr="shapebsdflabel_wd",
content_filter=content_filter,
title="Click on blobs that match an image (%s)" % attr,
description="This task involves clicking on images that match a blob next to the image.",
keywords="material,substance,shape,image,picture,classify,label,blob,match,appearance",
# frame_height=7500,
requirements={},
auto_add_hits=False, # settings.MTURK_SANDBOX,
)
def configure_experiments():
""" This function is automatically called by
the command ./manage.py mtconfigure """
# must be imported locally to avoid a circular import
from mturk.utils import configure_experiment
configure_experiment(
slug='rectify_continuous',
template_dir='normals/experiments',
module='normals.experiments',
examples_group_attr='shape',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.10'),
num_outputs_max=1,
contents_per_hit=10,
content_type_model=MaterialShape,
out_content_type_model=ShapeRectifiedNormalLabel,
out_content_attr='shape',
content_filter={
'invalid': False,
'planar': True,
'pixel_area__gt': 10 * Shape.MIN_PLANAR_AREA,
'num_vertices__gte': 4,
'correct': True,
'substance__isnull': False,
'substance__fail': False,
#'photo__whitebalanced': True,
'photo__scene_category_correct': True,
'photo__nonperspective': False,
},
title='Rotate an image in 3D',
description='Rotate a thumbtack so that a 3D image looks like it is facing the camera.',
keywords='flat,planar,appearance,polygon,shape,draw,image,picture,3D,graphics',
requirements={},
auto_add_hits=False, # settings.MTURK_SANDBOX,
)
configure_experiment(
slug='quality_rectify',
template_dir='normals/experiments',
module='normals.experiments',
examples_group_attr='shape',
version=1, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.04'),
num_outputs_max=5,
contents_per_hit=40,
content_type_model=ShapeRectifiedNormalLabel,
out_content_type_model=ShapeRectifiedNormalQuality,
out_content_attr='rectified_normal',
content_filter={
'invalid': False,
'shape__invalid': False,
'shape__planar': True,
#'shape__num_vertices__gte': 10,
'shape__pixel_area__gt': 10 * Shape.MIN_PLANAR_AREA,
#'automatic': True,
#'shape__correct': True,
#'shape__substance__isnull': False,
#'shape__photo__whitebalanced': True,
#'shape__photo__scene_category_correct': True,
},
title='Click on images that are correctly rotated',
description='This task involves clicking on images that are correctly rotated.',
keywords='material,substance,shape,image,picture,classify,label',
#frame_height=7500,
requirements={},
auto_add_hits=False, # settings.MTURK_SANDBOX,
)
def configure_experiments():
""" This function is automatically called by
the command ./manage.py mtconfigure """
from mturk.utils import configure_experiment
production = not settings.MTURK_SANDBOX
configure_experiment(
slug='intrinsic_opacity',
template_dir='intrinsic/experiments',
module='intrinsic.experiments',
version=2, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.05'),
num_outputs_max=5,
contents_per_hit=200, # if production else 50,
frame_height=668, # handle 1024x768 window sizes
content_type_model=IntrinsicPoint,
out_content_type_model=IntrinsicPointOpacityResponse,
out_content_attr='point',
content_filter={
#'photo__license__publishable': True,
#'photo__stylized': False,
#'photo__rotated': False,
#'photo__nonperspective': False,
#'photo__synthetic': True,
'photo__intrinsic_synthetic__multilayer_exr__isnull': False,
#'photo__scene_category_correct': True,
#'photo__num_intrinsic_points__gte': 20,
#'min_separation__lt': 0.05,
},
title='Click on mirrors and transparent surfaces',
description='Given points in images, your job is to decide which belong to mirrors or transparent objects.',
keywords='category,images,collection,image,picture,classify,color',
requirements={},
auto_add_hits=True,
has_tutorial=True,
)
# don't add photo 10386; it's somewhat ambiguous
# don't add photo 94610 -- it's all gloss and metal
excluded_photo_ids = [10386, 94610]
# hard-code examples here for now
equal_tests = IntrinsicPointComparison.objects.in_bulk(list(set([
961869, 961802, 961774, 961815, 961848, 961789, 961814, 961888, 961868,
961885, 961713, 961742, 961818, 961813, 961864, 961807, 961743, 961766,
961784, 961884, 961714, 999517, 999510, 1006818, 1006825, 1006838,
1006808, 1006820, 1006810, 1046422, 1046444, 1046369, 1046437, 1046352,
1046338, 1046425, 1046459, 929022, 929035, 929037, 929031, 929008,
928985, 929006, 929035, 2385973, 2386026, 2385959, 2386106, 2385974,
1605947, 1605914, 1605960, 1605989, 1605962, 1606032, 1605956, 1605931,
1605946, 1605988, 1606043, 1606044, 1118334, 1118351, 1118282, 1118347,
1118422, 1118329, 1118260, 1118410, 1118350, 1118443, 1118311, 1118306,
1118404, 1118362, 1118286, 1118405, 1118345, 1118287, 1118335, 1118367,
1118395, 1118390, 1118308, 1118400, 1118434, 1118301, 1118425, 1118439,
1118369, 1118415, 1118281, 1118288, 1118362, 2419322, 2419219, 2419210,
2419204, 956745, 956741, 956641, 928997, 928997, 928981, 1720063,
1720071, 1719956, 1719978, 1720041, 1720042, 1719928, 1720011, 1719905,
1719973, 1720059, 1719918, 1720012, 1720080, 1719915, 1720025, 1719974,
1719943, 1719908, 1719922, 1720044, 1720013, 1402497, 1402598, 1402545,
1719916, 1719958, 1720078, 1720628, 1720590, 1720606, 1720641, 1320564,
1320545, 1320547, 1320548, 2432036, 3708799, 3445549, 1118283, 1118258,
2508708, 951758, 951741, 951662, 951758, 951717, 951620, 951712, 951629,
]))).values()
for comparison in equal_tests:
if comparison.photo_id not in excluded_photo_ids:
comparison.darker = 'E'
comparison.darker_method = 'A'
comparison.save()
point1_tests = IntrinsicPointComparison.objects.in_bulk([
961811, 961865, 961759, 999516, 999490, 999491, 999583, 999500, 999564,
1046371, 2358770, 2358731, 1118331, 1118290, 1118263, 1118302, 2358729,
2358731, 2358695, 1180756, 1180681, 1180720, 1180701, 1180673, 976878,
976878, 999565, 1720053, 1719988, 1719951, 1402559, 1720077, 1720629,
1320551, 1320549, 1320627, 1320606, 1320622, 2431927, 2432017, 2431952,
2432053, 2431947, 2432055, 3708827, 3708860, 3708802, 3708808, 3708899,
3708874, 3708845, 3445595, 3445598, 3445566, 3445584, 3445498, 3708895,
3708908, 3708894, 3889277, 3889277, 3889172, 961827, 961769, 961891,
961725, 961744, 2369862, 2369839, 2369873, 2508687, 2508663, 2508695,
2508631
]).values()
for comparison in point1_tests:
if comparison.photo_id not in excluded_photo_ids:
comparison.darker = '1'
comparison.darker_method = 'A'
comparison.save()
point2_tests = IntrinsicPointComparison.objects.in_bulk([
999515, 999498, 999499, 999569, 999562, 1046453, 2358755, 2358801,
2358691, 1118365, 1118418, 2358714, 2358755, 956602, 956601, 1180669,
1180730, 976827, 976825, 1719954, 1720019, 1402521, 1402562, 1720565,
1320579, 2431925, 2431989, 2431923, 2432011, 3708773, 3708878, 3445472,
3445554, 3445586, 3445550, 3445609, 3445556, 3445546, 3889199, 3889278,
3889145, 2208252, 2369764, 2508705, 2508711
]).values()
for comparison in point2_tests:
if comparison.photo_id not in excluded_photo_ids:
comparison.darker = '2'
comparison.darker_method = 'A'
comparison.save()
test_contents = [t for t in (equal_tests + point1_tests + point2_tests)
if t.photo_id not in excluded_photo_ids]
configure_experiment(
slug='intrinsic_compare',
template_dir='intrinsic/experiments',
module='intrinsic.experiments',
version=2, # 2: intrinsic images, 1: original opensurfaces
reward=Decimal('0.06'),
num_outputs_max=5,
contents_per_hit=25, # if production else 2,
max_active_hits=10000 if production else 50,
frame_height=668, # handle 1024x768 window sizes
content_type_model=IntrinsicPointComparison,
out_content_type_model=IntrinsicPointComparisonResponse,
out_content_attr='comparison',
content_filter={
'point1__opaque': True,
'point2__opaque': True,
'photo__intrinsic_synthetic__multilayer_exr__isnull': False,
#'photo__license__publishable': True,
#'photo__stylized': False,
#'photo__rotated': False,
#'photo__nonperspective': False,
#'photo__synthetic': False,
#'photo__scene_category_correct': True,
#'point1__min_separation__lt': 0.05,
},
title='Compare colors in an image',
description='Given pairs of points in images, your job is to decide which of two points is darker.',
keywords='category,images,collection,image,picture,classify,color',
requirements={},
auto_add_hits=True,
has_tutorial=True,
test_contents=test_contents,
test_contents_per_assignment=5,
)
