train dataset: VOC 2012 + VOC 2007
test dataset: VOC 2007
test size: 544
test code: Faster rcnn (not use 07 metric)
test score threshold: 0.01

1. clone YOLO_v3 repository

   git clone https://github.com/qq751220449/yolov3_tensorflow.git

2. prepare data
   (1) create your datasets
   Create a new folder named data in the directory where the yolov3_tensorflow folder is located, and then create a new folder named VOC2007 in the data/.
   Then put datasets into data/VOC2007.

   The file structure is as follows:
   |--yolov3_tensorflow
   |--|--v1
   |--|--v2
   |--|--v3
   |--data
   |--|--VOC2007
   |--|--|--Annotations
   |--|--|--ImageSets
   |--|--|--|--Main
   |--|--|--JPEGImages
   (2) convert data format
   You should alter DATASET_PATH and PROJECT_PATHin config.py, for example:
   DATASET_PATH = /home/ice2019/yolov3_temsorflow/data
PROJECT_PATH = /home/ice2019/yolov3_temsorflow
   and then
   

   cd yolov3_tensorflow/utils
   python voc.py

3. prepare initial weights
   Download darknet53_448.weights firstly, put the initial weights into darknet2tf/, and then

   cd darknet2tf
   python3 convert_weights.py --weights_file=darknet53_448.weights --data_format=NHWC

4. train
   

   nohup python train.py &

5. test
   If you want to get a higher mAP, you can set the score threshold to 0.01、use multi scale test、flip test.
   If you want to use it in actual projects, or if you want speed, you can set the score threshold to 0.2.
   

   nohup python test.py --gpu=0 -t07 &

6. continue train
   You should alter Continue_To_Train and CHECKPOINT_FILEin config.py, for example:
   Continue_To_Train = True
CHECKPOINT_FILE = 'checkpoint/yolo.ckpt-99-0.7929'
   

7. convert checkpoint to pb file
   You should alter CKPT2PB_CKPT_FILE and CKPT2PB_PB_NAMEin config.py, for example:
   CKPT2PB_CKPT_FILE = 'checkpoint/yolo.ckpt-99-0.7929'
CKPT2PB_PB_NAME = 'frozen_model_20200103_99_7929.pb'
   

paper:

• YOLOv3: An Incremental Improvement
  
• Foca Loss for Dense Object Detection
  
• Group Normalization
  
• An Analysis of Scale Invariance in Object Detection - SNIP
  
• Deformable convolutional networks
  
• Scale-Aware Trident Networks for Object Detection
  
• Understanding the Effective Receptive Field in Deep Convolutional Neural Networks
  
• Bag of Freebies for Training Object Detection Neural Networks
  
• Generalized Intersection over Union: A Metric and A Loss for Bounding Box Regression
  
• Stonger-yolo
  

software

• Python3.6.5
  
• Numpy1.14.5
  
• Tensorflow.1.10.0
  
• Opencv3.4.1
  

hardware

• 8G 1080