wellton/TrainYourOwnYOLO-do-wellton
1
0
Fork 0
Código Issues Pull Requests Ações Pacotes Projetos Versões Wiki Atividade

feat: Simplify tiny YOLO training

Ver Código-Fonte 
Added some convenience features that make training the tiny model easier.
Esse commit está contido em:
AntonMu
2020-09-09 21:33:07 -07:00
pai c37417d44f
commit 63234a81ef
10 arquivos alterados com 82 adições e 30 exclusões
Baixar Arquivo de Patch Baixar Arquivo de Diff Expandir todos os arquivos Colapsar todos os arquivos
1_Image_Annotation/Convert_to_YOLO_format.py
+2 -2
Ver Arquivo
@@ -8,8 +8,8 @@ import argparse
def get_parent_dir(n=1):
""" returns the n-th parent dicrectory of the current
working directory """
"""returns the n-th parent dicrectory of the current
working directory"""
current_path = os.path.dirname(os.path.abspath(__file__))
for k in range(n):
current_path = os.path.dirname(current_path)
2_Training/Download_and_Convert_YOLO_weights.py
+27 -10
Ver Arquivo
@@ -27,29 +27,46 @@ if __name__ == "__main__":
help="Folder to download weights to. Default is " + download_folder,
)
parser.add_argument(
"--is_tiny",
default=False,
action="store_true",
help="Use the tiny Yolo version for better performance and less accuracy. Default is False.",
)
FLAGS = parser.parse_args()
if not os.path.isfile(os.path.join(download_folder, "yolov3.weights")):
print("\n", "Downloading Raw YOLOv3 Weights", "\n")
if not FLAGS.is_tiny:
weights_file = "yolov3.weights"
h5_file = "yolo.h5"
cfg_file = "yolov3.cfg"
# Original URL: https://pjreddie.com/media/files/yolov3.weights
gdrive_id = "1ENKguLZbkgvM8unU3Hq1BoFzoLeGWvE_"
else:
weights_file = "yolov3-tiny.weights"
h5_file = "yolo-tiny.h5"
cfg_file = "yolov3-tiny.cfg"
# Original URL: https://pjreddie.com/media/files/yolov3-tiny.weights
gdrive_id = "1mIEZthXBcEguMvuVAHKLXQX3mA1oZUuC"
if not os.path.isfile(os.path.join(download_folder, weights_file)):
print(f"\nDownloading Raw {weights_file}\n")
start = time.time()
call_string = " ".join(
[
"python",
download_script,
"1ENKguLZbkgvM8unU3Hq1BoFzoLeGWvE_",
os.path.join(download_folder, "yolov3.weights"),
gdrive_id,
os.path.join(download_folder, weights_file),
]
)
subprocess.call(call_string, shell=True)
end = time.time()
print(
"Downloaded Raw YOLOv3 Weights in {0:.1f} seconds".format(end - start), "\n"
)
print(f"Downloaded Raw {weights_file} in {end - start:.1f} seconds\n")
# Original URL: https://pjreddie.com/media/files/yolov3.weights
call_string = "python convert.py yolov3.cfg yolov3.weights yolo.h5"
call_string = f"python convert.py {cfg_file} {weights_file} {h5_file}"
subprocess.call(call_string, shell=True, cwd=download_folder)
2_Training/Train_YOLO.py
+9 -2
Ver Arquivo
@@ -158,8 +158,15 @@ if __name__ == "__main__":
class_names = get_classes(FLAGS.classes_file)
num_classes = len(class_names)
anchors = get_anchors(FLAGS.anchors_path)
weights_path = FLAGS.weights_path
if FLAGS.is_tiny and FLAGS.weights_path == weights_path:
weights_path = os.path.join(os.path.dirname(FLAGS.weights_path), "yolo-tiny.h5")
if FLAGS.is_tiny and FLAGS.anchors_path == anchors_path:
anchors_path = os.path.join(
os.path.dirname(FLAGS.anchors_path), "yolo-tiny_anchors.txt"
)
anchors = get_anchors(anchors_path)
input_shape = (416, 416) # multiple of 32, height, width
epoch1, epoch2 = FLAGS.epochs, FLAGS.epochs
2_Training/src/keras_yolo3/convert.py
+1 -1
Ver Arquivo
@@ -272,7 +272,7 @@ def _main(args):
remaining_weights = len(weights_file.read()) / 4
weights_file.close()
print(
"Read {} of {} from Darknet weights.".format(count, count + remaining_weights)
f"Read {count:0.0f} of {count + remaining_weights:0.0f} from Darknet weights."
)
if remaining_weights > 0:
print("Warning: {} unused weights".format(remaining_weights))
2_Training/src/keras_yolo3/yolo.py
+1
Ver Arquivo
@@ -21,6 +21,7 @@ import tensorflow.python.keras.backend as K
tf.disable_eager_execution()
class YOLO(object):
_defaults = {
"model_path": "model_data/yolo.h5",
3_Inference/Detector.py
+17 -3
Ver Arquivo
@@ -3,8 +3,8 @@ import sys
def get_parent_dir(n=1):
""" returns the n-th parent dicrectory of the current
working directory """
"""returns the n-th parent dicrectory of the current
working directory"""
current_path = os.path.dirname(os.path.abspath(__file__))
for k in range(n):
current_path = os.path.dirname(current_path)
@@ -28,6 +28,7 @@ import pandas as pd
import numpy as np
from Get_File_Paths import GetFileList
import random
from Train_Utils import get_anchors
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"
@@ -141,6 +142,13 @@ if __name__ == "__main__":
help='Specify the postfix for images with bounding boxes. Default is "_catface"',
)
parser.add_argument(
"--is_tiny",
default=False,
action="store_true",
help="Use the tiny Yolo version for better performance and less accuracy. Default is False.",
)
FLAGS = parser.parse_args()
save_img = not FLAGS.no_save_img
@@ -168,11 +176,17 @@ if __name__ == "__main__":
if not os.path.exists(output_path):
os.makedirs(output_path)
if FLAGS.is_tiny and FLAGS.anchors_path == anchors_path:
anchors_path = os.path.join(
os.path.dirname(FLAGS.anchors_path), "yolo-tiny_anchors.txt"
)
anchors = get_anchors(anchors_path)
# define YOLO detector
yolo = YOLO(
**{
"model_path": FLAGS.model_path,
"anchors_path": FLAGS.anchors_path,
"anchors_path": anchors_path,
"classes_path": FLAGS.classes_path,
"score": FLAGS.score,
"gpu_num": FLAGS.gpu_num,
Utils/Convert_Format.py
+20 -6
Ver Arquivo
@@ -9,7 +9,14 @@ from Get_File_Paths import GetFileList, ChangeToOtherMachine
def convert_vott_csv_to_yolo(
vott_df,
labeldict=dict(zip(["Cat_Face"], [0,])),
labeldict=dict(
zip(
["Cat_Face"],
[
0,
],
)
),
path="",
target_name="data_train.txt",
abs_path=False,
@@ -106,16 +113,16 @@ def crop_and_save(
):
"""Takes a vott_csv file with image names, labels and crop_boxes
and crops the images accordingly
Input csv file format:
image xmin ymin xmax ymax label
im.jpg 0 10 100 500 house
Parameters
----------
df : pd.Dataframe
df : pd.Dataframe
The input dataframe with file_names, bounding box info
and label
source_path : str
@@ -175,7 +182,14 @@ def crop_and_save(
if __name__ == "__main__":
# Prepare the houses dataset for YOLO
labeldict = dict(zip(["house"], [0,]))
labeldict = dict(
zip(
["house"],
[
0,
],
)
)
multi_df = pd.read_csv(
"C:/Users/Anton/Documents/Insight/eq/EQ_new/Train_Housing_detector/2/vott-csv-export/Housing_cropping-export.csv"
)
Utils/Get_File_Paths.py
+1 -1
Ver Arquivo
@@ -40,7 +40,7 @@ def ChangeToOtherMachine(filelist, repo="TrainYourOwnYOLO", remote_machine=""):
'/home/ubuntu/TrainYourOwnYOLO/Data/Street_View_Images/vulnerable/test.jpg'
Get's converted to
'C:/Users/Anton/TrainYourOwnYOLO/Data/Street_View_Images/vulnerable/test.jpg'
"""
Utils/Train_Utils.py
+3 -3
Ver Arquivo
@@ -9,8 +9,8 @@ import sys
def get_parent_dir(n=1):
""" returns the n-th parent dicrectory of the current
working directory """
"""returns the n-th parent dicrectory of the current
working directory"""
current_path = os.getcwd()
for k in range(n):
current_path = os.path.dirname(current_path)
@@ -221,7 +221,7 @@ def ChangeToOtherMachine(filelist, repo="TrainYourOwnYOLO", remote_machine=""):
'/home/ubuntu/TrainYourOwnYOLO/Data/Street_View_Images/vulnerable/test.jpg'
Get's converted to
'C:/Users/Anton/TrainYourOwnYOLO/Data/Street_View_Images/vulnerable/test.jpg'
"""
Utils/utils.py
+1 -2
Ver Arquivo
@@ -78,8 +78,7 @@ def load_extractor_model(model_name="InceptionV3", flavor=1):
flavor: int specifying the model variant and input_shape.
For InceptionV3, the map is {0: default, 1: 200*200, truncate last Inception block,
2: 200*200, truncate last 2 blocks, 3: 200*200, truncate last 3 blocks, 4: 200*200}
For VGG16, it only changes the input size, {0: 224 (default), 1: 128, 2: 64}.
"""
For VGG16, it only changes the input size, {0: 224 (default), 1: 128, 2: 64}."""
start = timer()
if model_name == "InceptionV3":
from keras.applications.inception_v3 import InceptionV3