def post(self):
try:
#Retrieve the image from http submit. It will be a blob object.
image = self.get_uploads()[0]
#Get info from blobfile uploaded
file_info = self.get_file_infos()[0]
#Create the serving url for the blob uploaded into GS (Google cloud storage)
serving_url = images.get_serving_url(blobstore.create_gs_key(file_info.gs_object_name))
#Fetch image dimensions. The last integer specifies how much to fetch 50000 is plenty.
data = blobstore.fetch_data(image.key(), 0, 50000)
img = images.Image(image_data = data)
width_to_height_ratio = float(img.width)/img.height
#Get name and description
name = self.request.get('name')
caption = self.request.get('caption')
#Now generate the database object and store it.
image_model = ImageModel(
name = name,
caption = caption,
blob_key = image.key(),
gs_object_name = file_info.gs_object_name,
serving_url = serving_url,
width_to_height_ratio = width_to_height_ratio)
image_model.put()
#Redirect to the upload site and tell it that it succeeded.
self.redirect('/admin/upload_image#success')
except:
#Redirect to upload site and tell that something went wrong.
self.redirect('/admin/upload_image#fail')
def main():
imageDataset = ImageDataSet(
"/Volumes/My Passport/abhishek/Datasets/Image Dataset/rvl-cdip/dataset"
)
model = ImageModel(8, 224, 224)
model.train_model(imageDataset, batch_size=50)
def receive_message():
if request.method == 'GET':
image_url = request.args["image"]
ImageModel.create(image_url)
prob, classes = model.predict(urllib.request.urlopen(image_url),
models)
max_index = np.argmax(prob)
max_probability = prob[max_index]
val_list = list(cat_to_name.values())
label = classes[max_index]
labels = []
for cl in classes:
labels.append(cat_to_name[str(val_list.index(cl))])
message = {"messages": [{"text": f'This is a picture of {labels[0]}'}]}
return make_response(jsonify(message))
def get(self):
redirect_if_not_admin(self)
images = ImageModel.query()
self.render("gallery_creation.html",
active="create_gallery",
images=images)
def get(self):
redirect_if_not_admin(self)
images = ImageModel.query()
self.render("image_admin.html",
active = "images",
images = images)
def get(self):
redirect_if_not_admin(self)
templates = Config.TEMPLATES
images = ImageModel.query()
main_pages = PageModel.query(PageModel.parent == None)
self.render("page_creation.html", active="create_page",
templates=templates,
main_pages=main_pages,
images=images)
def delete_image():
if request.method == "DELETE":
if "id" not in request.args:
return jsonify(msg="'id' cannot be left blank."), 400
try:
img_id = request.args.get("id")
image = ImageModel.find_by_id(img_id)
if image:
image.delete_from_database()
return jsonify(msg="Image with id '{}' has been successfully deleted.".format(img_id)), 200
return jsonify(msg="id '{}' does not exist.".format(img_id)), 400
except Exception as error:
return jsonify(msg="An error occured during image DELETE: {}".format(error)), 500
def get(self, curr_page_id="main"):
redirect_if_not_admin(self)
galleries = GalleryModel.query()
images = ImageModel.query()
curr_page = ""
if curr_page_id:
curr_page_query = GalleryModel.query(GalleryModel.gallery_id == curr_page_id)
curr_page = curr_page_query.get()
self.render("gallery_admin.html",
active="galleries",
galleries = galleries,
images = images,
curr_page = curr_page,
curr_page_id = curr_page_id)
def get(self, curr_page_id="main"):
redirect_if_not_admin(self)
pages = PageModel.query()
images = ImageModel.query()
curr_page = ""
if curr_page_id:
curr_page_query = PageModel.query(PageModel.page_id == curr_page_id)
curr_page = curr_page_query.get()
self.render('page_admin.html',
active="pages",
pages=pages,
templates = Config.TEMPLATES,
images = images,
main_pages = Config.MAIN_PAGES,
curr_page = curr_page,
curr_page_id = curr_page_id)
def Read(self, path, imgID):
imgName = self.path.split('/')[-1]
self.imagesModels[imgID] = ImageModel(self.path)
self.heights[imgID], self.weights[imgID] = self.imagesModels[
imgID].imgShape
if type(self.imagesModels[~imgID]) == type(...):
self.displayImage(self.imagesModels[imgID].imgByte,
self.inputImages[imgID])
logger.info(f"uplood Image{imgID + 1}: {imgName} ")
else:
if self.heights[1] != self.heights[0] or self.weights[
1] != self.weights[0]:
self.msg.setWindowTitle("Error in Image Size")
self.msg.setText("The 2 images don't have the same size.")
self.msg.setIcon(QMessageBox.Warning)
x = self.msg.exec_()
logger.error("Error: The two images don't have the same size.")
else:
self.displayImage(self.imagesModels[imgID].imgByte,
self.inputImages[imgID])
logger.info(f"uplood Image{imgID + 1}: {imgName} ")
# if isinstance(layer, Iterable):
# for i in layer:
# # _sublayers = list(i)
# print("%i", i)
# if isinstance(i, Iterable):
# for sublayer in _sublayers:
# print("sub_sublayer", sublayer)
# else:
# print(layer)
# --MAIN ------------------------------------------------------------------------------------------------------
if __name__ == '__main__':
# instantiate class to handle model
image_model = ImageModel(model_name='resnet50', dataset_size=2692)
# Initialize Image Module
# image_module = MyImageModule(dataset_size=100, batch_size=32)
image_module = MyImageModule(dataset_size=2692, batch_size=1)
image_module.setup()
# --- PREDICT RESULTS ---
# Get name and model used for testing
# name_model, inference_model = image_model.inference_model()
name_model = 'model-epoch=05-val_loss=0.36-weights7y3_unfreeze2.ckpt'
inference_model = image_model.load_model(name_model)
# print("Inference model:", inference_model)
print("Name:", name_model)
# Prediction with no tensors
# y_true, y_pred = predict(inference_model, image_module.test_dataloader())
def post_images():
if request.method == "POST":
#path to static
LOCAL_STATIC_PATH = 'static/'
UPLOADS_PATH = join(dirname(realpath(__file__)), LOCAL_STATIC_PATH)
DATABASE_SIZE = 100
## check to see if request body is filled
if "image" not in request.files:
return jsonify(msg="'image' cannot be left blank."), 400
elif "time_taken" not in request.form:
return jsonify(msg="'time_taken' cannot be left blank."), 400
elif "image_orientation" not in request.form:
return jsonify(msg="'image_orientation' cannot be left blank."), 400
# get the image
image = request.files.get("image")
time =request.form.get("time_taken")
orientation = request.form.get("image_orientation")
try:
seefoodWrapper.pollForReady()
# save the image to the server
image_name = secure_filename(image.filename)
path = UPLOADS_PATH + image_name
image.save(path)
# Seefood AI can take .bmp, .jpg, .png, etc. no need to convert. We can filter out images on app before
# they are sent to the server
confidences = seefoodWrapper.sendImage(path)
logger.write_info("routes.py:Got confidence ratings " + confidences[0] + " " +confidences[1])
# convert the image for server gallery
new_img = Image.open(path)
os.remove(path)
truncName = image_name.split(".")
newName = truncName[0] + '.png'
newNamePath = UPLOADS_PATH + newName
new_img.save(newNamePath, 'png')
logger.write_info("routes.py:Sucessfully re-saved image")
img_return_path = LOCAL_STATIC_PATH + newName
# ENFORCE DATABASE SIZE
all_images = ImageModel.query.all()
if len(all_images) == DATABASE_SIZE:
first_image = all_images[0]
first_image.delete_from_database()
imageModel = ImageModel(truncName[0], time, confidences[0], confidences[1], img_return_path, orientation )
logger.write_info("routes.py:Created imageModel")
imageModel.save_to_database()
logger.write_info("routes.py:Saved to database")
return jsonify(image=imageModel.json()), 201
except Exception as error:
logger.write_error(error)
return jsonify(msg="An error occured during image POST: {}".format(error)), 500
def stitchTwoImage(image_model1, image_model2):
stitch_img, raw_img = image_model1.image, image_model2.image
img1_gray = cv2.cvtColor(stitch_img, cv2.COLOR_BGR2GRAY)
img2_gray = cv2.cvtColor(raw_img, cv2.COLOR_BGR2GRAY)
# Feature matching
kp1, des1 = getFeature(img1_gray)
kp2, des2 = getFeature(img2_gray)
matches = featureMatching(des1, des2)
img_match = cv2.drawMatches(stitch_img, kp1, raw_img, kp2, matches, None)
drawMatchImage('match ' + image_model1.name + ' + ' + image_model2.name,
img_match)
# Calculate homography matrix
h, isHomographyGood = getHomographyMatrix(kp1, kp2, matches)
if not isHomographyGood:
print('[WARNING] Bad homography matrix, discard this data',
image_model2.name)
return image_model1, False
# Warping
print('Warping...')
start = time.time()
new_size = (raw_img.shape[1] + int(stitch_img.shape[1] * 1.5),
raw_img.shape[0] + int(stitch_img.shape[0] * 1.5))
# put the image to the central position of the result image
t = np.identity(3, np.float)
t[0, 2] = new_size[0] / 4
t[1, 2] = new_size[1] / 4
pad_raw = cv2.warpPerspective(raw_img,
t,
new_size,
borderMode=cv2.BORDER_REFLECT)
raw_mask = cv2.warpPerspective(raw_img, t, new_size)
pad_warp = cv2.warpPerspective(stitch_img,
t.dot(h),
new_size,
borderMode=cv2.BORDER_REFLECT)
warp_mask = cv2.warpPerspective(stitch_img, t.dot(h), new_size)
stitch_mask = np.logical_or(getMask(raw_mask), getMask(warp_mask))
stitch_mask = np.asarray(stitch_mask, dtype=np.uint8)
saveImage(
'h**o image {0} -> {1}'.format(image_model1.name, image_model2.name),
warp_mask, ImageModel.SAVE_HOMO)
print('--', time.time() - start, 's')
# Cut Padding
print('Cut padding...')
cut_pad_images = [pad_raw, pad_warp, raw_mask, warp_mask]
[cut_pad_raw, cut_pad_warp, cut_pad_raw_mask,
cut_pad_warp_mask] = cutPadding(cut_pad_images, stitch_mask)
if isAlreadyStitch(cut_pad_raw_mask, cut_pad_warp_mask):
print('Data {0} already stitched, so don\'t need to keep stitching.'.
format(image_model2.name))
return image_model1, True
final_mask = cv2.bitwise_or(getMask(cut_pad_raw_mask),
getMask(cut_pad_warp_mask))
final_mask = np.asarray(final_mask, dtype=np.uint8)
# Blending
print('blending images...')
blend_mask = (np.sum(cut_pad_raw_mask, axis=2) != 0).astype(np.float)
blend_img = multibandBlending(cut_pad_raw, cut_pad_warp, blend_mask)
blend_img = cv2.bitwise_and(blend_img, blend_img, mask=final_mask)
image_model_stitch = ImageModel(
image_model1.name + ' ' + image_model2.name, blend_img)
return image_model_stitch, True