def eval(data_directory="plates",
model_path="plates/models",
json_path="plates/json/detection_config.json",
iou_threshold=0.5,
object_threshold=0.3,
nms_threshold=0.5):
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory)
trainer.evaluateModel(model_path, json_path, iou_threshold,
object_threshold, nms_threshold)
def evaluate_yolov3(data_path, models_path, json_path, iou_thresh,
object_thresh, nms_thresh):
model = DetectionModelTrainer()
model.setModelTypeAsYOLOv3()
model.setDataDirectory(data_directory=data_path)
model.evaluateModel(model_path=models_path,
json_path=os.path.join(json_path,
"detection_config.json"),
iou_threshold=iou_thresh,
object_threshold=object_thresh,
nms_threshold=nms_thresh)
def train_yolo(lst_y, train_path="fs/training_yolo/", transfer_modelfile=""):
## setup
yolo = DetectionModelTrainer()
yolo.setModelTypeAsYOLOv3()
yolo.setDataDirectory(data_directory=train_path)
yolo.setTrainConfig(object_names_array=lst_y,
batch_size=4,
num_experiments=1,
train_from_pretrained_model=transfer_modelfile)
## train
print("--- training ---")
yolo.trainModel()
## evaluate
print("--- metrics ---")
metrics = yolo.evaluateModel(model_path=train_path + "models",
json_path=train_path +
"json/detection_config.json",
iou_threshold=0.5,
object_threshold=0.5,
nms_threshold=0.5)
print(metrics)
## laod model
print("--- loading model ---")
modelfile = os.listdir(train_path + "models/")[0]
return load_yolo(modelfile=train_path + "models/" + modelfile,
confjson=train_path + "/json/detection_config.json")
def evaluarModelo():
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory="2_RESPUESTA")
metrics = trainer.evaluateModel(model_path="models",
json_path="detection_config.json",
iou_threshold=0.5,
object_threshold=0.3,
nms_threshold=0.5)
print(metrics)
def main():
# Will we train?
if DOTRAIN:
# Use last trained model as base for transferlearning?
last_model = '' # Ohmy. This means None.
if DOTRANSFERLEARN:
last_model = MODEL_PATH
if DORESUMELAST:
last_model = get_last_model()
else:
if os.path.exists(os.path.join(DATA_PATH, 'cache')):
shutil.rmtree(os.path.join(DATA_PATH, 'cache'))
shutil.rmtree(os.path.join(DATA_PATH, 'json'))
shutil.rmtree(os.path.join(DATA_PATH, 'logs'))
shutil.rmtree(os.path.join(DATA_PATH, 'models'))
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory=DATA_PATH)
trainer.setTrainConfig(object_names_array=OBJECT_NAMES,
batch_size=BATCHCOUNT,
num_experiments=EXPIREMENTSCOUNT,
train_from_pretrained_model=last_model)
trainer.trainModel()
# Will we evaluate?
if DOEVALUATE:
# Show evaluations for the last model
metrics = trainer.evaluateModel(model_path=get_last_model(),
json_path=os.path.join(
DATA_PATH, 'json',
'detection_config.json'),
iou_threshold=.5,
object_threshold=.3,
nms_threshold=.5)
print(metrics)
from imageai.Detection.Custom import DetectionModelTrainer
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory="datasets")
trainer.evaluateModel(model_path="datasets/models",
json_path="datasets/json/detection_config.json",
iou_threshold=0.5,
object_threshold=0.3,
nms_threshold=0.5)
from imageai.Detection.Custom import DetectionModelTrainer
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory="smoker")
trainer.evaluateModel(
model_path="smoker/models/detection_model-ex-143--loss-0007.288.h5",
json_path="smoker/json/detection_config.json",
iou_threshold=0.1,
object_threshold=0.35,
nms_threshold=0.1)
# download JSON file via https://github.com/OlafenwaMoses/ImageAI/releases/download/essential-v4/detection_config.json
# download detection model via https://github.com/OlafenwaMoses/ImageAI/releases/download/essential-v4/hololens-ex-60--loss-2.76.h5
"""
SAMPLE RESULT
Model File: hololens_detection_model-ex-09--loss-4.01.h5
Using IoU : 0.5
Using Object Threshold : 0.3
Using Non-Maximum Suppression : 0.5
hololens: 0.9613
mAP: 0.9613
===============================
"""
from imageai.Detection.Custom import DetectionModelTrainer trainer = DetectionModelTrainer() trainer.setModelTypeAsYOLOv3() trainer.setDataDirectory(data_directory="qr") metrics = trainer.evaluateModel(model_path="detection_model-ex-001--loss-0000.000.h5", json_path="detection_config.json", iou_threshold=0.5, object_threshold=0.3, nms_threshold=0.5)
trainer.setDataDirectory(data_directory="dataset_yolo_v3/")
trainer.setTrainConfig(object_names_array=["corn", "weed"],
batch_size=2,
num_experiments=200,
train_from_pretrained_model="pretrained-yolov3.h5")
trainer.trainModel()
"""
Evaluate model
- evaluate how every model checkpoint performs via its mAP.
"""
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory="dataset_yolo_v3")
metrics = trainer.evaluateModel(
model_path="dataset_yolo_v3/models",
json_path="dataset_yolo_v3/json/detection_config.json",
iou_threshold=0.5,
object_threshold=0.9,
nms_threshold=0.5)
"""
Use model for detection
"""
detector = CustomObjectDetection()
detector.setModelTypeAsYOLOv3()
detector.setModelPath(
"dataset_yolo_v3/models/detection_model-ex-005--loss-0014.777.h5")
detector.setJsonPath("dataset_yolo_v3/json/detection_config.json")
detector.loadModel()
detections = detector.detectObjectsFromImage(
input_image="test.jpg",
output_image_path="ima-detected.jpg",
import os
import glob
from imageai.Detection.Custom import DetectionModelTrainer
execution_path = os.getcwd()
models_path = os.path.join(execution_path, "doge-identification/models/")
data_path = os.path.join(execution_path, "doge-identification/")
# This will force tensorflow to run on the cpu
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
list_of_files = glob.glob(models_path + "*.h5")
latest_model_path = max(list_of_files, key=os.path.getctime)
path, newest_model = os.path.split(latest_model_path)
print("Newest Model: ", newest_model)
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory=data_path)
metrics = trainer.evaluateModel(
model_path=latest_model_path,
json_path=os.path.join(execution_path, "doge-identification/json/detection_config.json"),
iou_threshold=0.5,
object_threshold=0.3,
nms_threshold=0.5)
from imageai.Detection.Custom import DetectionModelTrainer
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory="Vier_Onderdelen_Wit")
trainer.setTrainConfig(
object_names_array=["Schoepen_groot", "Schijf_zilver", "Kom", "Rotor"],
batch_size=4,
num_experiments=100,
train_from_pretrained_model="pretrained-yolov3.h5")
trainer.trainModel()
trainer.setDataDirectory(data_directory="Vier_Onderdelen_Wit")
evaluate = trainer.evaluateModel(
model_path="Vier_Onderdelen_Wit/models",
json_path="Vier_Onderdelen_Wit/json/detection_config.json",
iou_threshold=0.5,
object_threshold=0.3,
nms_threshold=0.5)
x = []
for i in range(len(evaluate)):
map = evaluate[i]['map']
experiment = evaluate[i]['model_file']
x.append((map, experiment))
x.sort()
text_file = open("maps_vier_onderdelen_wit.txt", "wt")
for i in x:
e = text_file.writelines(str(i) + "\n")
text_file.close()
# -*- coding: utf-8 -*- #Run this in colab or in a jupyter notebook """Prerequisites""" !pip3 install tensorflow-gpu==1.13.1 !pip3 install imageai --upgrade !wget https://github.com/OlafenwaMoses/ImageAI/releases/download/essential-v4/pretrained-yolov3.h5 """Change data_directory and object_names_array appropriately""" from imageai.Detection.Custom import DetectionModelTrainer trainer = DetectionModelTrainer() trainer.setModelTypeAsYOLOv3() trainer.setDataDirectory(data_directory="/content/drive/My Drive/Football/") trainer.setTrainConfig(object_names_array=["Bundesliga", "EPL", "LaLiga"], batch_size=4, num_experiments=7, train_from_pretrained_model="pretrained-yolov3.h5") trainer.trainModel() """Pick the model with maximum mAP""" from imageai.Detection.Custom import DetectionModelTrainer trainer = DetectionModelTrainer() trainer.setModelTypeAsYOLOv3() trainer.setDataDirectory(data_directory="/content/drive/My Drive/Football") trainer.evaluateModel(model_path="/content/drive/My Drive/Football/models", json_path="/content/drive/My Drive/Football/json/detection_config.json", iou_threshold=0.5, object_threshold=0.3, nms_threshold=0.5)
from imageai.Detection.Custom import DetectionModelTrainer
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(
data_directory="/content/drive/My Drive/project data sets/data_cleaned")
trainer.evaluateModel(
model_path="/content/drive/My Drive/project data sets/data_cleaned/models",
json_path=
"/content/drive/My Drive/project data sets/data_cleaned/json/detection_config.json",
iou_threshold=0.5,
object_threshold=0.3,
nms_threshold=0.5)
from imageai.Detection.Custom import DetectionModelTrainer
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory="../data/")
metrics = trainer.evaluateModel(
model_path="../models/detection_model-ex-015--loss-0004.540.h5",
json_path="../json/detection_config.json",
iou_threshold=0.5,
object_threshold=0.3,
nms_threshold=0.5,
)
print(metrics)
from imageai.Detection.Custom import DetectionModelTrainer
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(
data_directory="C:/Users/ABC/Desktop/New folder/TDAI/Data/hololens")
trainer.evaluateModel(
model_path="C:/Users/ABC/Desktop/New folder/TDAI/Data/hololens/models",
json_path=
"C:/Users/ABC/Desktop/New folder/TDAI/Data/hololens/json/detection_config.json",
iou_threshold=0.5,
object_threshold=0.3,
nms_threshold=0.5)
from imageai.Detection.Custom import DetectionModelTrainer
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory="SymDescribeImages")
metrics = trainer.evaluateModel(
model_path="SymDescribeImages/models",
json_path="SymDescribeImages/json/detection_config.json",
iou_threshold=0.5,
object_threshold=0.3,
nms_threshold=0.5)
print(metrics)
from imageai.Detection.Custom import DetectionModelTrainer
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory="characters")
trainer.setTrainConfig(object_names_array=[
'aeolianDune', 'beach', 'crevasseRiverBedLagDeposit',
'crevasseRiverChannelBar', 'distalBar', 'distributaryChannelFill',
'distributaryMouthBar', 'floodPlain', 'floodSwamp',
'interdistributaryEstuary', 'majorRiverBedLagDeposit',
'majorRiverChannelBar', 'meanderPointBar', 'mouthBar',
'proximalDeepseaFan', 'underwaterDistributaryChannel',
'underwaterNaturalLevee'
],
batch_size=2,
num_experiments=10,
train_from_pretrained_model="pretrained-yolov3.h5")
trainer.trainModel()
metrics = trainer.evaluateModel(
model_path="characters/models",
json_path="characters/json/detection_config.json",
iou_threshold=0.5,
object_threshold=0.3,
nms_threshold=0.5)
print(metrics)
'''
After running the training file, object_detection_custom_training.py,
Run this file to evaluate the generated model from training
'''
from imageai.Detection.Custom import DetectionModelTrainer
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory="hololens")
trainer.evaluateModel(
model_path="hololens/models", # where model is
json_path="hololens/json/detection_config.json",
iou_threshold=
0.5, # minimum desired inersect over union of the Mean Average Precision
object_threshold=0.3, #desired minimum class score for the mAP computation
nms_threshold=0.5
) #desired Non-maximum suppression for the mAP computation.
from imageai.Detection.Custom import DetectionModelTrainer
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory="Dataset2")
trainer.setTrainConfig(object_names_array=["Vehicle"], batch_size=4, num_experiments=20, train_from_pretrained_model="pretrained-yolov3.h5")
trainer.trainModel()
# Evaluation
from imageai.Detection.Custom import DetectionModelTrainer
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory="Dataset2")
metrics = trainer.evaluateModel(model_path="Dataset2/models", json_path="Dataset2/json/detection_config.json", iou_threshold=0.5, object_threshold=0.3, nms_threshold=0.5)
# Detection
from imageai.Detection.Custom import CustomObjectDetection
def deet(fold,best):
detector = CustomObjectDetection()
detector.setModelTypeAsYOLOv3()
detector.setModelPath("Dataset{}/models/{}.h5".format(fold,best))
detector.setJsonPath("Dataset{}/json/detection_config.json".format(fold))
detector.loadModel()
for ikl in [1,2,3,4,5,6]:
detections = detector.detectObjectsFromImage("Dataset{}/test/T{}.jpg".format(fold,ikl), output_image_path="Dataset{}/test/T{}_detected.jpg".format(fold,ikl),nms_treshold=0.1, minimum_percentage_probability=30)
from imageai.Detection.Custom import DetectionModelTrainer
trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory="hololens")
trainer.evaluateModel(model_path="hololens-ex-60--loss-2.76.h5",
json_path="detection_config.json",
iou_threshold=0.5,
object_threshold=0.3,
nms_threshold=0.5)
# download JSON file via https://github.com/OlafenwaMoses/ImageAI/releases/download/essential-v4/detection_config.json
# download detection model via https://github.com/OlafenwaMoses/ImageAI/releases/download/essential-v4/hololens-ex-60--loss-2.76.h5
"""
SAMPLE RESULT
Model File: hololens_detection_model-ex-09--loss-4.01.h5
Using IoU : 0.5
Using Object Threshold : 0.3
Using Non-Maximum Suppression : 0.5
hololens: 0.9613
mAP: 0.9613
===============================
"""
