val_coco_dir = ""
val_img_dir = ""
val_set_dir = system["val_monk_img_dir"]
else:
val_root_dir = system["val_coco_root_dir"]
val_coco_dir = system["val_coco_coco_dir"]
val_img_dir = system["val_coco_img_dir"]
val_set_dir = system["val_coco_set_dir"]
system["epochs"] = int(system["epochs"])
system["val_interval"] = int(system["val_interval"])
system["lr"] = float(system["lr"])
gtf = Detector()
gtf.Train_Dataset(root_dir,
coco_dir,
img_dir,
set_dir,
batch_size=system["batch_size"],
image_size=system["image_size"],
use_gpu=system["use_gpu"])
if (system["val_data"] == "yes"):
gtf.Val_Dataset(val_root_dir, val_coco_dir, val_img_dir, val_set_dir)
tmp = system["devices"].split(",")
gpu_devices = []
for i in range(len(tmp)):
# In[8]: from train_detector import Detector # In[9]: # pwd # In[10]: gtf = Detector(); # In[11]: root_dir = "."; coco_dir = "coco_dataset_3class"; img_dir = "."; set_dir = "Images"; # In[12]: gtf.Train_Dataset(root_dir, coco_dir, img_dir, set_dir, batch_size=8, image_size=512, use_gpu=True)
val_coco_dir = ""
val_img_dir = ""
val_set_dir = system["val_monk_img_dir"]
else:
val_root_dir = system["val_coco_root_dir"]
val_coco_dir = system["val_coco_coco_dir"]
val_img_dir = system["val_coco_img_dir"]
val_set_dir = system["val_coco_set_dir"]
system["epochs"] = int(system["epochs"])
system["val_interval"] = int(system["val_interval"])
system["lr"] = float(system["lr"])
gtf = Detector()
gtf.Train_Dataset(root_dir,
coco_dir,
img_dir,
set_dir,
batch_size=system["batch_size"],
use_gpu=system["use_gpu"])
if (system["val_data"] == "yes"):
gtf.Val_Dataset(val_root_dir, val_coco_dir, val_img_dir, val_set_dir)
tmp = system["devices"].split(",")
gpu_devices = []
for i in range(len(tmp)):
gpu_devices.append(int(tmp[i]))
# Check TF version
import tensorflow as tf
print(tf.__version__)
import os
import sys
sys.path.append("MONK/Monk_Object_Detection/13_tf_obj_2/lib/")
from train_detector import Detector
gtf = Detector()
print(gtf.list_models())
train_img_dir = "COCO_CREATION/results/Train/images"
train_anno_dir = "COCO_CREATION/results/Train/annotations"
class_list_file = "COCO_CREATION/pascal-voc-classes.txt"
gtf.set_train_dataset(train_img_dir,
train_anno_dir,
class_list_file,
batch_size=24,
trainval_split=0.8)
## Output dir
output_dir = os.path.join("data_tfrecord")
gtf.create_tfrecord(data_output_dir=output_dir)
copy_to_root()
train_num_img=sum([1 for i in os.listdir("/content/Root_Dir/Coco_style/images/Train")])
valid_num_img=sum([1 for i in os.listdir("/content/Root_Dir/Coco_style/images/Val")])
print("There are {} train images and {} valid images.".format(train_num_img,valid_num_img))
#!pip install efficientnet_pytorch
#!pip install tensorboardx
#https://github.com/Tessellate-Imaging/Monk_Object_Detection/blob/master/example_notebooks/5_pytorch_retinanet/Train%20Resnet18%20-%20With%20validation%20Dataset.ipynb
import os
os.chdir("/content/Computer-vision-and-Drones-Thesis/models/Modified Monk Retinanet/lib")
from train_detector import Detector
model = Detector();
os.chdir("/content/")
troot_dir = "Root_Dir";
tcoco_dir = "Coco_style";
timg_dir = "images";
tset_dir = "Train";
vroot_dir = "Root_Dir";
vcoco_dir = "Coco_style";
vimg_dir = "images";
vset_dir = "Val";
val_img_dir = "";
val_set_dir = system["val_monk_img_dir"];
else:
val_root_dir = system["val_coco_root_dir"];
val_coco_dir = system["val_coco_coco_dir"];
val_img_dir = system["val_coco_img_dir"];
val_set_dir = system["val_coco_set_dir"];
gtf = Detector();
gtf.Train_Dataset(root_dir, coco_dir, img_dir, set_dir,
batch_size=system["batch_size"],
num_workers=4)
if(system["val_data"] == "yes"):
gtf.Val_Dataset(val_root_dir, val_coco_dir, val_img_dir, val_set_dir)
gtf.Model(model_name=system["model"]);
gtf.Hyper_Params(lr=system["lr"],
val_root_dir = system["val_coco_root_dir"] + "/" + system[
"val_coco_coco_dir"]
val_img_dir = system["val_coco_img_dir"] + "/" + system[
"val_coco_set_dir"]
val_anno_dir = labels_dir
else:
val_root_dir = system["val_yolo_root_dir"]
val_img_dir = system["val_yolo_img_dir"]
val_anno_dir = system["val_yolo_anno_dir"]
val_classes_file = system["val_yolo_classes_file"]
from train_detector import Detector
gtf = Detector()
gtf.set_train_dataset(root_dir + "/" + img_dir,
root_dir + "/" + anno_dir,
root_dir + "/" + classes_file,
batch_size=system["batch_size"],
img_size=system["img_size"],
cache_images=system["cache_images"])
if (system["val_data"] == "yes"):
gtf.set_val_dataset(val_root_dir + "/" + val_img_dir,
val_root_dir + "/" + val_anno_dir)
gtf.set_model(model_name=system["model"])
gtf.set_hyperparams(optimizer=system["optimizer"],