tonyzhang1231 / transferlearningforobjectdetection

TransferLearningForObjectDetection

Transfer Learning for Object Detection Task, using InceptionV3 and YOLO in tensorflow

The goal is to predict all the bounding boxes that containing an object and the category of the object for each bounding box.

We used InceptionV3 as the base feature extractor and trained a new classifier, we implemented the loss function proposed in YOLO's paper as the loss function.

Clone The Git

mkdir project_dir && cd project_dir

git clone https://github.com/tonyzhang1231/TransferLearningForObjectDetection

Get The Pascal VOC Data

mkdir image_dir && cd image_dir

wget https://pjreddie.com/media/files/VOCtrainval_11-May-2012.tar

wget https://pjreddie.com/media/files/VOCtrainval_06-Nov-2007.tar

wget https://pjreddie.com/media/files/VOCtest_06-Nov-2007.tar

tar xf VOCtrainval_11-May-2012.tar

tar xf VOCtrainval_06-Nov-2007.tar

tar xf VOCtest_06-Nov-2007.tar

Generate Labels for VOC

We need to generate the label files that our model uses. we need a .txt file for each image with a line for each ground truth object in the image that looks like:

<object-class> <x> <y> <width> <height>

x, y are the coordinate of the bounding box. all x, y, w, h are between 0 and 1, relative to the size of the image.

Copy the voc_label.py to image_dir and run this file

cp project_dir/voc_label.py image_dir/

cd image_dir && python voc_label.py

In the image_dir, run the following command to get the train, val and test data

cat 2007_train.txt 2012_*.txt > train.txt

cat 2007_val.txt > validation.txt

cat 2007_test.txt > test.txt

Change the settings in the run_retrain_InceptionV3_YOLO.sh file

Change the IMAGE_DIR and result_store_dir and anything else if you want

Retrain the model

sh run_retrain_InceptionV3_YOLO.sh

Requirements

• tensorflow-gpu==1.4.1
• python 2.7.12

highly recommend to use python environment

reference

• darknet/yolo https://pjreddie.com/darknet/yolo/
• darkflow https://github.com/thtrieu/darkflow

transferlearningforobjectdetection's Issues

How to use the new model?

Hi, thanks for your sharing.
I've retrained the model by following your instructions, but how to use this new model to do OBJECT DETECTION jobs? Could you please show me some examples?
Thanks.