def test_deep_same(self):
finder = DeepFinder()
# pattern matching is not perfect
finder.params["find"]["similarity"].value = 0.95
matches = finder.find(Pattern('cat'), Image('coco_cat'))
# verify match accuracy
self.assertEqual(len(matches), 1)
self.assertAlmostEqual(matches[0].x, 90, delta=5)
self.assertAlmostEqual(matches[0].y, 345, delta=5)
self.assertAlmostEqual(matches[0].width, 515, delta=5)
self.assertAlmostEqual(matches[0].height, 805, delta=5)
# verify dumped files count and names
dumps = self._verify_and_get_dumps(6)
self._verify_dumped_images('cat', 'coco_cat', dumps, "deep")
hotmaps = sorted(self._get_matches_in('.*hotmap.*', dumps))
self.assertEqual(len(hotmaps), 3)
for i, hotmap in enumerate(hotmaps):
if i == 0:
self.assertIn('3hotmap', hotmap)
# report achieved similarity in the end of the filename
self.assertRegex(hotmap, ".*-\d\.\d+.*")
else:
self.assertIn('%sf' % i, hotmap)
self.assertTrue(os.path.isfile(os.path.join(self.logpath, hotmap)))
finder.configure_backend("tensorflow", "deep")
with self.assertRaises(ImportError):
finder.synchronize_backend()
with self.assertRaises(NotImplementedError):
finder.find(Pattern('cat'), Image('coco_cat'))
def test_cascade_viewport(self):
finder = CascadeFinder()
matches = finder.find(Pattern('n_ibs.xml'), Image('h_ibs_viewport'))
self.assertEqual(len(matches), 1)
# the original needle image was 150x150 with larger white margins
self.assertAlmostEqual(matches[0].x, 20, delta=5)
self.assertAlmostEqual(matches[0].y, 20, delta=5)
# near square shape is due to the positive images used for training
self.assertAlmostEqual(matches[0].width, 250, delta=10)
self.assertAlmostEqual(matches[0].height, 250, delta=10)
def test_cascade_nomatch(self):
finder = CascadeFinder()
# no similarty parameter is supported - this is a binary match case
finder.params["find"]["similarity"].value = 0.0
matches = finder.find(Pattern('n_ibs.xml'), Image('all_shapes'))
# verify match accuracy
self.assertEqual(len(matches), 0)
# verify dumped files count and names
dumps = self._verify_and_get_dumps(4)
self._verify_dumped_images('n_ibs', 'all_shapes', dumps, "cascade")
self._verify_single_hotmap(dumps, "cascade")
def test_deep_cache(self):
finder = DeepFinder(synchronize=False)
finder.params["deep"]["arch"].value = "fasterrcnn_resnet50_fpn"
finder.synchronize_backend()
matches = finder.find(Pattern('cat'), Image('coco_cat'))
self.assertEqual(len(matches), 1)
self.assertEqual(len(finder._cache.keys()), 1)
self.assertEqual(finder._cache[finder.params["deep"]["arch"].value],
finder.net)
matches = finder.find(Pattern('cat'), Image('coco_cat'))
self.assertEqual(len(matches), 1)
self.assertEqual(len(finder._cache.keys()), 1)
self.assertEqual(finder._cache[finder.params["deep"]["arch"].value],
finder.net)
finder.params["deep"]["arch"].value = "maskrcnn_resnet50_fpn"
finder.synchronize_backend()
matches = finder.find(Pattern('cat'), Image('coco_cat'))
self.assertEqual(len(matches), 1)
self.assertEqual(len(finder._cache.keys()), 2)
self.assertEqual(finder._cache[finder.params["deep"]["arch"].value],
finder.net)
def test_cascade_same(self):
finder = CascadeFinder()
# no similarty parameter is supported - this is a binary match case
finder.params["find"]["similarity"].value = 0.0
matches = finder.find(Pattern('shape_blue_circle.xml'),
Image('all_shapes'))
# verify match accuracy
self.assertEqual(len(matches), 1)
self.assertAlmostEqual(matches[0].x, 104, delta=5)
self.assertAlmostEqual(matches[0].y, 10, delta=5)
self.assertAlmostEqual(matches[0].width, 165, delta=10)
self.assertAlmostEqual(matches[0].height, 151, delta=10)
# verify dumped files count and names
dumps = self._verify_and_get_dumps(4)
self._verify_dumped_images('shape_blue_circle', 'all_shapes', dumps,
"cascade")
self._verify_single_hotmap(dumps, "cascade")
def test_deep_nomatch(self):
finder = DeepFinder()
finder.params["find"]["similarity"].value = 0.25
matches = finder.find(Pattern('cat'), Image('all_shapes'))
# verify match accuracy
self.assertEqual(len(matches), 0)
# verify dumped files count and names
dumps = self._verify_and_get_dumps(6)
self._verify_dumped_images('cat', 'all_shapes', dumps, "deep")
hotmaps = sorted(self._get_matches_in('.*hotmap.*', dumps))
self.assertEqual(len(hotmaps), 3)
for i, hotmap in enumerate(hotmaps):
if i == 0:
self.assertIn('3hotmap', hotmap)
# report achieved similarity in the end of the filename
self.assertRegex(hotmap, ".*-\d\.\d+.*")
else:
self.assertIn('%sf' % i, hotmap)
self.assertTrue(os.path.isfile(os.path.join(self.logpath, hotmap)))
finder.net = FasterRCNN(backbone, num_classes=len(dataset.classes))
elif "mask" in finder.params["deep"]["arch"].value:
from torchvision.models.detection.mask_rcnn import MaskRCNN
finder.net = MaskRCNN(backbone, num_classes=len(dataset.classes))
# TODO: eventually support keypoint R-CNN if it shows to be promising
#elif "keypoint" in finder.params["deep"]["arch"].value:
# from torchvision.models.detection.keypoint_rcnn import KeypointRCNN
# finder.net = KeypointRCNN(backbone, num_classes=len(dataset.classes))
else:
raise ValueError(
f'Invalid choice of architecture: {finder.params["deep"]["arch"].value}'
)
finder.net.to(device)
# Train and test the network
for epoch in range(1, hyperparams["epochs"] + 1):
train(epoch, finder.net, train_loader, device, hyperparams)
test(epoch, finder.net, test_loader, device, hyperparams)
# Evaluate the trained network on a single test sample with image logging
NEEDLE = Pattern("1")
HAYSTACK = Image('some_buttons')
matches = finder.find(NEEDLE, HAYSTACK)
# Final cleanup steps
if REMOVE_LOGPATH:
shutil.rmtree(LOGPATH)
GlobalConfig.image_logging_level = logging.ERROR
GlobalConfig.image_logging_destination = "./imglog"
GlobalConfig.image_logging_step_width = 3
def test_cascade_rotation(self):
finder = CascadeFinder()
matches = finder.find(Pattern('n_ibs.xml'), Image('h_ibs_rotated'))