def test_detect_frcnn_long(self):
for im, name, labels, boxes in utils.tf_record_iterator(
f"{BASE_DIR}/frcnn_records/test/annotated.record"):
image_methods.show(im,
use_log=False,
annotations=zip(boxes, labels))
answer = image_methods.detect_comet(im, do_show=True)
def test_pipeline(self):
x = utils.image_iterator(FITS_FOLDER)
for _ in range(160):
scanfiles = next(x)
print(scanfiles[0])
ims = scanfiles[0]
ims = [image_methods.get_image_data(im)[0] for im in ims]
origs = ims.copy()
image_methods.show_many(np.array(origs))
preprod = [image_methods.preprocess(im, resize=False) for im in ims]
resized = [image_methods.resize_image(im, 512) for im in preprod]
image_methods.show_many(np.array(origs), row_2=np.array(resized))
composite = image_methods.get_composite_image(resized)
image_methods.show(composite, use_log=False)
image_methods.detect_comet(composite,
do_show=True,
im_type="composite")
def test_untrained_comet_detection(self):
for images in utils.image_iterator(f"{BASE_DIR}/not_trained_comets"):
fits = images[0]
arrs = [image_methods.get_image_data(f)[0] for f in fits]
preprod = [
image_methods.preprocess(arr, resize=False) for arr in arrs
]
resized = [image_methods.resize_image(pre, 512) for pre in preprod]
comp = image_methods.get_composite_image(resized)
image_methods.show(comp)
answer = image_methods.detect_comet(preprod,
do_show=True,
im_type="composite")
print(answer)
def test_detect_3frcnn_long(self):
acc, count = (0, 0)
for im, name, labels, boxes in utils.tf_record_iterator(
f"{BASE_DIR}/data/band_3/c/test.record"):
print(name)
image_methods.show(im,
use_log=False,
annotations=zip(boxes, labels))
answer = image_methods.detect_comet(im,
do_show=True,
im_type="band_3")
acc += answer == ("comet" in labels)
count += 1
print(f"accuracy: {acc / count}")
def test_detect_comet_frcnn(self):
folders = ["comet", "messier", "not_comet"]
ans = []
for folder in folders:
orig_file = f"{BASE_DIR}/data1/{folder}/original.npy"
origs = np.load(orig_file)
#if there are 4 bands, get the 4th band
if origs.shape[0] == 4:
origs = origs[3]
rand = np.random.choice(np.r_[:len(origs)])
imarray = origs[rand]
answer = image_methods.detect_comet(imarray, do_show=True)
ans.append(answer)
self.assertEqual(ans[0] is not False, True)
self.assertEqual(ans[1:], [False] * len(ans[1:]))
def detect(fits_data,
im_type="all_bands",
model="faster_rcnn_inception_v2_coco",
classes=["comet"],
do_show="False"):
"""
fits_data (list or string) if list, list of fits files.
If string, path to single fits file.
im_type (string) band_3, all_bands, or composite. What
type of neural net to use for detection
model (string) directory in models. Probably don't want
to change this. can download additional models
from tf object detection model zoo
classes: Which classes the neural net you want to use is
trained on. For example, if your neural net was
trained on classes named "roses" and "dandelions",
classes would look like: ["roses", "dandelions"]
do_show: Whether or not to show the image in a matplotlib
popup window
---
RETURNS:
imarray (array) the image, preprocessed and resized
boxes (list) list of bounding box coordinates
scores (list) list of confidences for each bounding box
classes (list) list of classes that each bounding box was classified as
"""
classes_string = "".join([x[0] for x in classes])
if isinstance(fits_data, list):
images = [get_image_data(f)[0] for f in fits_data]
else:
images = [get_image_data(fits_data)[0]]
images = [preprocess(im, resize=False) for im in images]
imarray, boxes, scores, classes = detect_comet(
images,
do_show=do_show,
im_type=im_type,
model=model,
classes_string=classes_string)
return imarray, boxes, scores, classes