YOLOV3-detecting-red-blood-cell

this is a yolov3 model that can detecting red-blood-cell

Introduction

A Keras implementation of YOLOv3 (Tensorflow backend) inspired by https://github.com/qqwweee/keras-yolo3

final results

how to start

(1) just test model using pretrained weight

1、Download YOLOv3 weights from YOLO website.
2、wget https://pjreddie.com/media/files/yolov3.weights or directly open this link on windows to download  
3、Convert the Darknet YOLO model to a Keras model: python convert.py yolov3.cfg yolov3.weights model_data/yolo.h5
4、Run YOLO detection.  
5、modify yolo.py as         
    "model_path": 'model_data/yolo.h5',  
    "anchors_path": 'model_data/yolo_anchors.txt',  
    "classes_path": 'model_data/coco_classes.txt',  
6、python yolo_video.py [OPTIONS...] --image, for image detection mode(you will input image_path later on the console)
7、python yolo_video.py [video_path] [output_path (optional)], for video detection

(2) train on your own data

1、download red-blood-cell dataset: https://github.com/cosmicad/dataset
2、Convert the Darknet YOLO model to a Keras model: 
    python convert.py -w darknet53.cfg darknet53.weights model_data/darknet53_weights.h5
3、transfer annotation files into a txt file
    just modify wyk_parase_anno.py--change "anno_dir" and "img_dir" to your dataset dir
    results format just like:
        One row for one image;
        Row format: image_file_path box1 box2 ... boxN;
        Box format: x_min,y_min,x_max,y_max,class_id (no space).
        (For VOC dataset, try python voc_annotation.py)
        Here is an example:
        path/to/img1.jpg 50,100,150,200,0 30,50,200,120,3
        path/to/img2.jpg 120,300,250,600,2
        ...
4、if you wana use the pretrained weights , download here : https://pjreddie.com/media/files/darknet53.conv.74
   use the pretrained weights and you will get a better performance in a little dataset.  
   \if U do not wana use the pretrained weights, just modify wyk_train_redcell.py - load_pretrained=False.
5、change to your own path, modify wyk_train_redcell.py-
    annotation_path = 'redcell_anno_wyk.txt'
    log_dir = 'logs_redcell/'
    classes_path = 'model_data/redcell_classes.txt'
    anchors_path = 'model_data/yolo_anchors.txt'
 6、run pyhton wyk_train_redcell.py and later you will see the weights in loggig dir

(3)for more dataset to enjoy

Hand detection : http://cvrr.ucsd.edu/vivachallenge/index.php/hands/hand-detection/
Self-driving Car detection : http://cocodataset.org/#detections-challenge2017
Kangaroo detection : https://github.com/experiencor/kangaroo
Raccon detection : https://github.com/experiencor/raccoon_dataset

tips

these datasets are less images, so using pretrained will get a better performance than you train the model from the beginning.

platform

Keras 2.2.4
Tensorflow-gpu 1.12
windows 10