This repository was forked from https://github.com/nilboy/tensorflow-yolo and trained on own data.

• tensorflow >=1.0
• numpy
• OpenCV >=3.0

• added data augmentation in the training process (yolo/dataset/text_dataset.py: line 161 )
• modified the demo.py for multi-object detection with non-maximum-suppression (demo_image.py, demo_video.py)
• added the evaluation code to compute the precision and recall (demo_dir_for_pr.py, evaluation.py).
• added the code prepared for mAP computation (for mAP evaluation, please refer to https://github.com/Cartucho/mAP )
• trained on own data with the pretrained tiny yolo model: https://drive.google.com/file/d/0B-yiAeTLLamRekxqVE01Yi1RRlk/view?usp=sharing

1. download the dataset
   the url of the dataset is in ./data/dataset_url.txt

2. prepare the data
   generate the annotation (a text file) of the images for training, the code is in ./tools/preprocess_stick_cup_pen.py

3. train the network with .cfg file

   python tools/train.py -c conf/train.cfg

• demo_image.py: test an image
• demo_video.py: test a video
• demo_dir_for_pr.py: test the images in specified directory (this code will generate a text file for precision and recall computation)
• demo_dir_for_mAP.py: test the images in specified directory (this code will generate a text file for mAP computation)
• evaluation.py: compute the precisiona and recall
• mAP computation: refer to https://github.com/Cartucho/mAP