Skip to content

Stick-To/Object-Detection-Tensorflow

Repository files navigation

Object-Detection-API-Tensorflow

Features

Every model is implemented in only one file!

Models

Yolo2

Yolo3

SSD

RetinaNet

RefineDet

Light Head Rcnn

PFPNet

CenterNet

FCOS

Train your own data

Train your own data

1. TFRecord generation

  1. voc format dataset

  2. fill in utils.voc_classname_encoder.py

  3. run utils.test_voc_utils.py

2. config online image augmentor

fill in dict 'image_augmentor_config' in test-model.py

see utils.image_augmentor.py for details

see https://github.com/Stick-To/Online_Image_Augmentor_tensorflow for details

3. config model

fill in dict 'config' in test-model.py

4. Train

run test-model.py

The pre-trained vgg_16.ckpt could be downloaded from http://download.tensorflow.org/models/vgg_16_2016_08_28.tar.gz

5. Test

run annotated code in test-model.py

6. ImageNet pretraining

see utils.tfrecord_imagenet_utils.py

7. different conv backone

https://github.com/Stick-To/Deep_Conv_Backone

8. Instantiation of result

corresponding repository in https://github.com/Stick-To

Experimental Environment

python3.7 tensorflow1.13