def test_random_transform(annotations):
test_model = deepforest.deepforest()
test_model.config["random_transform"] = True
classes_file = utilities.create_classes(annotations)
arg_list = utilities.format_args(annotations, classes_file,
test_model.config)
assert "--random-transform" in arg_list
def config():
print("Configuring tfrecord tests")
config = {}
config["patch_size"] = 200
config["patch_overlap"] = 0.05
config["annotations_xml"] = get_data("OSBS_029.xml")
config["rgb_dir"] = "data"
config["annotations_file"] = "tests/data/OSBS_029.csv"
config["path_to_raster"] =get_data("OSBS_029.tif")
config["image-min-side"] = 800
config["backbone"] = "resnet50"
#Create a clean config test data
annotations = utilities.xml_to_annotations(xml_path=config["annotations_xml"])
annotations.to_csv("tests/data/tfrecords_OSBS_029.csv",index=False)
annotations_file = preprocess.split_raster(path_to_raster=config["path_to_raster"],
annotations_file="tests/data/tfrecords_OSBS_029.csv",
base_dir= "tests/data/",
patch_size=config["patch_size"],
patch_overlap=config["patch_overlap"])
annotations_file.to_csv("tests/data/testfile_tfrecords.csv", index=False,header=False)
class_file = utilities.create_classes("tests/data/testfile_tfrecords.csv")
return config
def evaluate_generator(self,
annotations,
comet_experiment=None,
iou_threshold=0.5,
max_detections=200):
""" Evaluate prediction model using a csv fit_generator
Args:
annotations (str): Path to csv label file, labels are in the format -> path/to/image.png,x1,y1,x2,y2,class_name
iou_threshold(float): IoU Threshold to count for a positive detection (defaults to 0.5)
max_detections (int): Maximum number of bounding box predictions
comet_experiment(object): A comet experiment class objects to track
Return:
mAP: Mean average precision of the evaluated data
"""
#Format args for CSV generator
classes_file = utilities.create_classes(annotations)
arg_list = utilities.format_args(annotations, classes_file, self.config)
args = parse_args(arg_list)
#create generator
validation_generator = CSVGenerator(
args.annotations,
args.classes,
image_min_side=args.image_min_side,
image_max_side=args.image_max_side,
config=args.config,
shuffle_groups=False,
)
average_precisions = evaluate(validation_generator,
self.prediction_model,
iou_threshold=iou_threshold,
score_threshold=args.score_threshold,
max_detections=max_detections,
save_path=args.save_path,
comet_experiment=comet_experiment)
# print evaluation
total_instances = []
precisions = []
for label, (average_precision, num_annotations) in average_precisions.items():
print('{:.0f} instances of class'.format(num_annotations),
validation_generator.label_to_name(label),
'with average precision: {:.4f}'.format(average_precision))
total_instances.append(num_annotations)
precisions.append(average_precision)
if sum(total_instances) == 0:
print('No test instances found.')
return
print('mAP using the weighted average of precisions among classes: {:.4f}'.format(
sum([a * b for a, b in zip(total_instances, precisions)]) /
sum(total_instances)))
mAP = sum(precisions) / sum(x > 0 for x in total_instances)
print('mAP: {:.4f}'.format(mAP))
return mAP
def from_retinanet(annotations_csv):
model = deepforest.deepforest()
model.use_release()
"""use the keras retinanet source to create detections"""
### Keras retinanet
# Format args for CSV generator
classes_file = utilities.create_classes(annotations_csv)
arg_list = utilities.format_args(annotations_csv, classes_file,
model.config)
args = parse_args(arg_list)
# create generator
validation_generator = csv_generator.CSVGenerator(
args.annotations,
args.classes,
image_min_side=args.image_min_side,
image_max_side=args.image_max_side,
config=args.config,
shuffle_groups=False,
)
all_detections = _get_detections(validation_generator,
model.prediction_model,
score_threshold=args.score_threshold,
max_detections=100)
all_annotations = _get_annotations(validation_generator)
return all_detections, all_annotations, validation_generator
def test_format_args_steps(annotations, config):
classes_file = utilities.create_classes(annotations)
arg_list = utilities.format_args(annotations,classes_file, config, images_per_epoch=2)
assert isinstance(arg_list, list)
#A bit ugly, but since its a list, what is the argument after --steps to assert
steps_position = np.where(["--steps" in x for x in arg_list])[0][0] + 1
assert arg_list[steps_position] == '2'
def test_retrain_release(annotations, release_model):
release_model.config["epochs"] = 1
release_model.config["save-snapshot"] = False
release_model.config["steps"] = 1
assert release_model.config["weights"] == release_model.weights
#test that it gets passed to retinanet
classes_file = utilities.create_classes(annotations)
arg_list = utilities.format_args(annotations,
classes_file,
release_model.config,
images_per_epoch=1)
strs = ["--weights" == x for x in arg_list]
index = np.where(strs)[0][0] + 1
assert arg_list[index] == release_model.weights
def train(self,
annotations,
input_type="fit_generator",
list_of_tfrecords=None,
comet_experiment=None,
images_per_epoch=None):
"""Train a deep learning tree detection model using keras-retinanet.
This is the main entry point for training a new model based on either
existing weights or scratch.
Args:
annotations (str): Path to csv label file,
labels are in the format -> path/to/image.png,x1,y1,x2,y2,class_name
input_type: "fit_generator" or "tfrecord"
list_of_tfrecords: Ignored if input_type != "tfrecord",
list of tf records to process
comet_experiment: A comet ml object to log images. Optional.
images_per_epoch: number of images to override default config
of images in annotations file / batch size. Useful for debug
Returns:
model (object): A trained keras model
prediction model: with bbox nms
trained model: without nms
"""
# Test if there is a new classes file in case # of classes has changed.
self.classes_file = utilities.create_classes(annotations)
self.read_classes()
arg_list = utilities.format_args(annotations, self.classes_file, self.config,
images_per_epoch)
print("Training retinanet with the following args {}".format(arg_list))
# Train model
self.model, self.prediction_model, self.training_model = retinanet_train(
forest_object=self,
args=arg_list,
input_type=input_type,
list_of_tfrecords=list_of_tfrecords,
comet_experiment=comet_experiment)
def predict_generator(self,
annotations,
comet_experiment=None,
iou_threshold=0.5,
max_detections=200,
return_plot=False,
color=None):
"""Predict bounding boxes for a model using a csv fit_generator
Args:
annotations (str): Path to csv label file, labels are in the
format -> path/to/image.png,x1,y1,x2,y2,class_name
comet_experiment(object): A comet experiment class objects to track
color: rgb color for the box annotations if return_plot is True e.g. (255,140,0) is orange.
return_plot: Whether to return prediction boxes (False) or Images (True). If True, files will be written to current working directory if model.config["save_path"] is not defined.
Return:
boxes_output: If return_plot=False, a pandas dataframe of bounding boxes
for each image in the annotations file
None: If return_plot is True, images are written to save_dir as a side effect.
"""
# Format args for CSV generator
classes_file = utilities.create_classes(annotations)
arg_list = utilities.format_args(annotations, classes_file, self.config)
args = parse_args(arg_list)
# create generator
generator = CSVGenerator(
args.annotations,
args.classes,
image_min_side=args.image_min_side,
image_max_side=args.image_max_side,
config=args.config,
shuffle_groups=False,
)
if self.prediction_model:
boxes_output = []
# For each image, gather predictions
for i in range(generator.size()):
# pass image as path
plot_name = generator.image_names[i]
image_path = os.path.join(generator.base_dir, plot_name)
result = self.predict_image(image_path,
return_plot=return_plot,
score_threshold=args.score_threshold,
color=color)
if return_plot:
if not self.config["save_path"]:
print("model.config['save_path'] is None,"
"saving images to current working directory")
save_path = "."
else:
save_path = self.config["save_path"]
# Save image
fname = os.path.join(save_path, plot_name)
cv2.imwrite(fname, result)
continue
else:
# Turn boxes to pandas frame and save output
box_df = pd.DataFrame(result)
# use only plot name, not extension
box_df["plot_name"] = os.path.splitext(plot_name)[0]
boxes_output.append(box_df)
else:
raise ValueError(
"No prediction model loaded. Either load a retinanet from file, "
"download the latest release or train a new model")
if return_plot:
return None
else:
# if boxes, name columns and return box data
boxes_output = pd.concat(boxes_output)
boxes_output.columns = [
"xmin", "ymin", "xmax", "ymax", "score", "label", "plot_name"
]
boxes_output = boxes_output.reindex(
columns=["plot_name", "xmin", "ymin", "xmax", "ymax", "score", "label"])
return boxes_output
def predict_generator(self,
annotations,
comet_experiment=None,
iou_threshold=0.5,
max_detections=200):
"""Predict bounding boxes for a model using a csv fit_generator
Args:
annotations (str): Path to csv label file, labels are in the format -> path/to/image.jpg,x1,y1,x2,y2,class_name
iou_threshold(float): IoU Threshold to count for a positive detection (defaults to 0.5)
max_detections (int): Maximum number of bounding box predictions
comet_experiment(object): A comet experiment class objects to track
Return:
boxes_output: a pandas dataframe of bounding boxes for each image in the annotations file
"""
#Format args for CSV generator
classes_file = utilities.create_classes(annotations)
arg_list = utilities.format_args(annotations, classes_file,
self.config)
args = parse_args(arg_list)
#create generator
generator = CSVGenerator(
args.annotations,
args.classes,
image_min_side=args.image_min_side,
image_max_side=args.image_max_side,
config=args.config,
shuffle_groups=False,
)
if self.prediction_model:
boxes_output = []
#For each image, gather predictions
for i in range(generator.size()):
#pass image as path
plot_name = generator.image_names[i]
image_path = os.path.join(generator.base_dir, plot_name)
boxes = self.predict_image(
image_path,
return_plot=False,
score_threshold=args.score_threshold)
#Turn to pandas frame and save output
box_df = pd.DataFrame(boxes)
#use only plot name, not extension
box_df["plot_name"] = os.path.splitext(plot_name)[0]
boxes_output.append(box_df)
else:
raise ValueError(
"No prediction model loaded. Either load a retinanet from file, download the latest release or train a new model"
)
#name columns and return box data
boxes_output = pd.concat(boxes_output)
boxes_output.columns = [
"xmin", "ymin", "xmax", "ymax", "score", "label", "plot_name"
]
boxes_output = boxes_output.reindex(columns=[
"plot_name", "xmin", "ymin", "xmax", "ymax", "score", "label"
])
return boxes_output
def test_format_args(annotations, config):
classes_file = utilities.create_classes(annotations)
arg_list = utilities.format_args(annotations, classes_file, config)
assert isinstance(arg_list, list)
def test_create_classes(annotations):
classes_file = utilities.create_classes(annotations_file=annotations)
assert os.path.exists(classes_file)
#Prepare data
#Load annotations file
annotations = pd.read_csv(
BASE_PATH + "pretraining/crops/pretraining.csv",
names=["image_path", "xmin", "ymin", "xmax", "ymax", "label"])
#Select a set of n image
annotations = annotations[annotations.image_path ==
"2019_DELA_5_423000_3601000_image_0.jpg"].copy()
#Generate tfrecords
annotations_file = BASE_PATH + "pretraining/crops/test.csv"
annotations.to_csv(annotations_file, header=False, index=False)
class_file = utilities.create_classes(annotations_file)
tfrecords_path = tfrecords.create_tfrecords(annotations_file,
class_file,
size=1)
print("Created {} tfrecords: {}".format(len(tfrecords_path), tfrecords_path))
inputs, targets = tfrecords.create_tensors(tfrecords_path)
#### Fit generator ##
comet_experiment = Experiment(api_key="ypQZhYfs3nSyKzOfz13iuJpj2",
project_name="deepforest",
workspace="bw4sz")
comet_experiment.log_parameter("Type", "testing")
comet_experiment.log_parameter("input_type", "fit_generator")
def generate_hand_annotations(DEBUG, BASE_PATH, FILEPATH, SIZE, config, dask_client):
#Generate tfrecords
dirname = "hand_annotations/"
annotations_file = BASE_PATH + dirname + "crops/hand_annotations.csv"
class_file = utilities.create_classes(annotations_file)
if DEBUG:
tfrecords.create_tfrecords(annotations_file=annotations_file,
class_file=class_file,
image_min_side=config["image-min-side"],
backbone_model=config["backbone"],
size=SIZE,
savedir=FILEPATH + dirname + "tfrecords/")
else:
#Collect annotation files for each tile
annotations_file= BASE_PATH + dirname + "crops/hand_annotations.csv"
df = pd.read_csv(annotations_file, names=["image_path","xmin","ymin","xmax","ymax","label"])
#enforce dtype, as there might be errors
df.xmin = df.xmin.astype(pd.Int64Dtype())
df.ymin = df.ymin.astype(pd.Int64Dtype())
df.xmax = df.xmax.astype(pd.Int64Dtype())
df.ymax = df.ymax.astype(pd.Int64Dtype())
#Randomize rows
df = df.sample(frac=1)
#split pandas frame into chunks
images = df.image_path.unique()
indices = np.arange(len(images))
size = 500
chunk_list = [ ]
#Split dataframe into chunks of images and write to file
for i in range(ceil(len(indices) / size)):
image_indices = indices[i * size:(i * size) + size]
selected_images = images[image_indices]
split_frame = df[df.image_path.isin(selected_images)]
filename = BASE_PATH + dirname + "crops/hand_annotations{}.csv".format(i)
split_frame.to_csv(filename, header=False,index=False)
chunk_list.append(filename)
print(" Created {} files to create tfrecords".format(len(chunk_list)))
#Apply create tfrecords to each
futures = dask_client.map(
tfrecords.create_tfrecords,
chunk_list,
class_file=class_file,
image_min_side=config["image-min-side"],
backbone_model=config["backbone"],
size=SIZE,
savedir=FILEPATH + dirname + "tfrecords/")
wait(futures)
for future in futures:
try:
local_annotations = future.result()
except Exception as e:
print("future {} failed with {}".format(future, e))