def test(batch_size, seed_size, filters, context):
mx.random.seed(int(time.time()))
net_g = Generator(filters)
net_g.load_parameters("model/toy_gan.generator.params", ctx=context)
seeds = mx.nd.random_normal(shape=(batch_size, seed_size, 1, 1), ctx=context)
imgs = net_g(seeds)
for i in range(imgs.shape[0]):
plt.subplot(imgs.shape[0] // 8 + 1, 8, i + 1)
visualize(imgs[i])
plt.show()
def view_dataset(self, mode='train'):
train_pairs, valid_pairs = dataset.prepare_data_CANCER()
if mode == 'train':
train_augmentors = self.train_augmentors()
ds = dataset.DatasetSerial(train_pairs,
shape_augs=iaa.Sequential(train_augmentors[0]),
input_augs=iaa.Sequential(train_augmentors[1]))
else:
infer_augmentors = self.infer_augmentors()
ds = dataset.DatasetSerial(valid_pairs,
shape_augs=iaa.Sequential(infer_augmentors))
dataset.visualize(ds, 1)
return
def view_dataset(self, mode='train'):
train_pairs, valid_pairs = getattr(dataset, ('prepare_%s_data' % self.dataset))()
if mode == 'train':
train_augmentors = self.train_augmentors()
ds = dataset.DatasetSerial(train_pairs, has_aux=False,
shape_augs=iaa.Sequential(train_augmentors[0]),
input_augs=iaa.Sequential(train_augmentors[1]))
else:
infer_augmentors = self.infer_augmentors() # HACK
ds = dataset.DatasetSerial(valid_pairs, has_aux=False,
shape_augs=iaa.Sequential(infer_augmentors)[0])
dataset.visualize(ds, 4)
return
def view_dataset(self, mode='train'):
train_pairs, valid_pairs = dataset.prepare_PAIP2020_PANDA(
self.fold_idx)
if mode == 'train':
train_augmentors = self.train_augmentors()
ds = dataset.DatasetSerial(train_pairs, self.tile_size,
self.num_tile, True)
else:
infer_augmentors = self.infer_augmentors() # HACK
ds = dataset.DatasetSerial(valid_pairs, self.tile_size,
self.num_tile, False)
dataset.visualize(ds, 1)
return
def detect_plate(wpod, vocab, ocr, raw, dims, threshold, plt_hw, beam,
beam_size, context):
h = raw.shape[0]
w = raw.shape[1]
f = min(288 * max(h, w) / min(h, w), 608) / min(h, w)
ts = time.time()
img = mx.image.imresize(raw,
int(w * f) + (0 if w % 16 == 0 else 16 - w % 16),
int(h * f) + (0 if h % 16 == 0 else 16 - h % 16))
x = color_normalize(img).transpose((2, 0, 1)).expand_dims(0)
y = wpod(x.as_in_context(context))
probs = y[0, :, :, 0]
affines = y[0, :, :, 2:]
labels = plate_labels(img, probs, affines, dims, 16, threshold)
plates = reconstruct_plates(raw, [pts for pts, _ in labels],
(plt_hw[1], plt_hw[0]))
print("wpod profiling: %f" % (time.time() - ts))
plt.subplot(math.ceil((len(plates) + 2) / 2), 2, 1)
visualize(img, [(pts.reshape((-1)).asnumpy().tolist(), str(prob))
for pts, prob in labels])
plt.subplot(math.ceil((len(plates) + 2) / 2), 2, 2)
visualize(probs > threshold)
for i, plate in enumerate(plates):
plt.subplot(math.ceil((len(plates) + 2) / 2), 2, i + 3)
visualize(plate)
print("plate[%d]:" % i)
recognize_plate(vocab, ocr, plate, beam, beam_size, context)
plt.show()
def test(dataset, batch_size, filters, context):
datasets = {
"facades": True,
"cityscapes": False,
"maps": False,
"edges2shoes": False,
"edges2handbags": False
}
mx.random.seed(int(time.time()))
print("Loading dataset...", flush=True)
validating_set = load_dataset(dataset,
"val",
batch_size,
is_reversed=datasets[dataset])
net_g = UnetGenerator(3, filters)
net_g.load_parameters("model/{}.generator.params".format(dataset),
ctx=context)
print("Testing...", flush=True)
for batch in validating_set:
real_in = batch.data[0].as_in_context(context)
real_out = batch.data[1].as_in_context(context)
fake_out = net_g(real_in)
for i in range(batch_size):
plt.subplot(3, batch_size, i + 1)
visualize(real_in[i])
plt.subplot(3, batch_size, i + batch_size + 1)
visualize(real_out[i])
plt.subplot(3, batch_size, i + batch_size * 2 + 1)
visualize(fake_out[i])
plt.show()
return A.Compose(
train_transform,
p=base_proba,
)
def get_validation_augmentation(self):
test_transform = []
return A.Compose(test_transform)
def get_preprocessing(self, preprocessing_fn):
_transform = [A.Lambda(image=preprocessing_fn)]
return A.Compose(_transform)
if __name__ == "__main__":
from dataset import Dataset, visualize
data_path = "data"
dataset = Dataset(
images_dir=os.path.join(data_path, "test"),
masks_dir=os.path.join(data_path, "train_annotations"),
classes=["bulk"],
augmentations=Augmentor.get_training_augmentation(),
preprocessing=None,
)
image, mask = dataset[5]
visualize(image=image, bulk_mask=mask[..., 0].squeeze())
def predict_dataset(
self,
dataset: Dataset,
output_folder: str = None,
if_blobs: bool = False,
review: bool = False,
binary_threshold: float = None,
):
elements_count = len(dataset)
for i in range(elements_count):
image, mask_true = dataset[i]
# mask_pred is an np array of shape (256, 256, 1)
# with values in range(0,1)
mask_pred = (self.predict(image) * 255).astype(np.uint8)
# binary mask if needed
if binary_threshold:
mask_pred = ((mask_pred > binary_threshold) * 255).astype(
np.uint8)
# find blobs on prediction
if if_blobs:
detector = SkimageBlobDetector(images=None)
# detector = OpenCVBlobDetector(images=None)
try:
with open(f"best_blob_params_{detector.name}.pickle",
"rb") as f:
ext_params = pickle.load(f)
except Exception as e:
print(e)
ext_params = {}
try:
noise_threshold = ext_params.pop("noise_threshold")
# filter little dark gray noise
mask_pred[mask_pred < noise_threshold] = 0
min_radius = ext_params.pop("min_radius")
max_radius = ext_params.pop("max_radius")
except Exception as e:
print("Params not found", e)
min_radius = 0
max_radius = np.inf
# invert colors only for detector
keypoints = self.detect_blobs(
image=255 -
mask_pred if detector.name == "opencv" else mask_pred,
detector=detector,
ext_params=ext_params,
)
keypoints_filtered = detector.filter_keypoints(
keypoints=keypoints,
min_radius=min_radius,
max_radius=max_radius,
)
mask_pred = detector.draw_blobs(
image=mask_pred,
keypoints=keypoints_filtered,
)
# obtaining filename for saving if needed
(
base_name,
file_format,
) = os.path.split(dataset.masks_fps[i])[-1].split(".")
save_name = os.path.join(
output_folder if output_folder else "",
f"{base_name}"
f"{'_' + str(i) if self.if_crop else ''}"
f"{'_' + str(len(keypoints_filtered)) if if_blobs else ''}"
f"{'_blobs' if if_blobs else ''}."
f"{file_format}",
)
print(save_name)
# mode for saving image
if review:
visualize(
save_name=save_name,
image=denormalize(image.squeeze()),
mask_true=mask_true,
mask_pred=mask_pred,
)
else:
self.save_image(
mask_pred,
filename=save_name,
)
testset = NuclieTestDataset(test_data_dir)
test_loader = DataLoader(dataset=testset,
drop_last=False,
sampler=SequentialSampler(testset),
batch_size=1)
model = model.cuda()
with torch.no_grad():
for step, (image) in enumerate(test_loader):
print(image.shape)
imgshow = image.clone()
imgshow = imgshow.squeeze()
imgshow = imgshow.permute(1, 2, 0).numpy()
imgshow = imgshow.astype(int)
visualize(imgshow)
image = image.cuda()
pred_mask_1 = model.forward(image)
print(pred_mask_1.shape)
mask_img_1 = mask_convert(pred_mask_1)
### round values ###
#mask_img_1 = np.round(mask_img_1)
#mask_img_2 = np.round(mask_img_2)
mask_img_1 = cv.normalize(mask_img_1, None, 0, 1, cv.NORM_MINMAX)
### normalize values to 0-1 ###
#print(mask_img_1)
mask_img_1 = np.round(mask_img_1)
visualize(mask_img_1)
break
linear_svm_model = svm.LinearSVC()
linear_svm_model.fit(features, labels)
test_data = dataset.get_test_data()
features, labels = dataset.get_imgs_feature(test_data, hog, hog_params)
results = linear_svm_model.predict(features)
util.precision_and_recall_evaluate(results, labels)
pos, neg = dataset.get_predict_data()
for i in range(5):
rects = util.detect_person(pos[i], hog, hog_params, linear_svm_model)
rects = util.non_max_suppression_fast(np.array(rects), 0.5)
print(pos[i])
print(rects)
dataset.visualize(pos[i], rects)
# # imgs = []
# # img_paths = []
# rects = util.detect_person(
# 'dataset/test/pos/crop_000001.jpg', hog, hog_params, linear_svm_model)
# print(np.array(rects))
# rect = util.non_max_suppression_fast(np.array(rects), 0.5)
# print(rect)
# # images = glob.glob('inria/sliding/*.jpg')
# # for image in images:
# # feature = hog.compute(cv2.imread(image),winStride,padding,locations)
# # feature = feature.reshape(-1)
# # imgs.append(feature)
# # img_paths.append(image)
