The object-recognition-yolov3 from stefanos19

part 1. Introduction

Implementation of YOLO v3 object detector in Tensorflow (TF-Slim). This repository is inspired by Paweł Kapica and Kiril Cvetkov . The full details are in this paper. In this project we cover several segments as follows:

YOLO v3 architecture
Weights converter (util for exporting loaded COCO weights as TF checkpoint)
Basic working demo
Non max suppression on the both GPU and CPU is supported
Training pipeline
Compute COCO mAP

YOLO paper is quick hard to understand, along side that paper. This repo enables you to have a quick understanding of YOLO Algorithmn.

part 2. Quick start

Clone this file

$ git clone https://github.com/YunYang1994/tensorflow-yolov3.git

You are supposed to install some dependencies before getting out hands with these codes.

$ cd tensorflow-yolov3
$ pip install -r ./docs/requirements.txt

Exporting loaded COCO weights as TF checkpoint(yolov3.ckpt) and frozen graph (yolov3_gpu_nms.pb) . If you don't have yolov3.weights. Download and put it in the dir ./checkpoint

$ python convert_weight.py --convert --freeze

Then you will get some .pb files in the dir ./checkpoint, and run the demo script

$ python nms_demo.py
$ python video_demo.py # if use camera, set video_path = 0

part 3. Train on your own dataset

Three files are required as follows:

dataset.txt:

xxx/xxx.jpg 18.19 6.32 424.13 421.83 20 323.86 2.65 640.0 421.94 20 
xxx/xxx.jpg 55.38 132.63 519.84 380.4 16
# image_path x_min y_min x_max y_max class_id  x_min y_min ... class_id

anchors.txt

0.10,0.13, 0.16,0.30, 0.33,0.23, 0.40,0.61, 0.62,0.45, 0.69,0.59, 0.76,0.60,  0.86,0.68,  0.91,0.76

class.names

person
bicycle
car
...
toothbrush

3.1 train raccoon dataset

To help you understand my training process, I made this training-pipline demo. raccoon dataset has only one class, I have prepared a shell script in the './scripts which enables you to get data and train it ! Finally python quick_test.py , here I strongly recommend you to set iou_thresh = 0.5, score_thresh=0.3.

how to train it ?

$ sh scripts/make_raccoon_tfrecords.sh
$ python show_input_image.py               # show your input image (optional)
$ python kmeans.py                         # get prior anchors and rescale the values to the range [0,1]
$ python convert_weight.py --convert       # get pretrained weights
$ python quick_train.py
$ tensorboard --logdir ./data

As you can see in the tensorboard, if your dataset is too small or you train for too long, the model starts to overfit and learn patterns from training data that does not generalize to the test data.

how to test and evaluate it ?

$ python convert_weight.py -cf ./checkpoint/yolov3.ckpt-2500 -nc 1 -ap ./data/raccoon_anchors.txt --freeze
$ python quick_test.py
$ python evaluate.py

if you are still unfamiliar with training pipline, you can join here to discuss with us.

raccoon-181.jpg	raccoon-55.jpg

3.2 train other dataset

Download VOC-2007 trainval and test data

$ wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtrainval_06-Nov-2007.tar
$ wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tar

Download COCO trainval and test data

$ wget http://images.cocodataset.org/zips/train2017.zip
$ wget http://images.cocodataset.org/annotations/annotations_trainval2017.zip
$ wget http://images.cocodataset.org/zips/test2017.zip
$ wget http://images.cocodataset.org/annotations/image_info_test2017.zip

part 4. Why it is so magical ?

YOLO stands for You Only Look Once. It's an object detector that uses features learned by a deep convolutional neural network to detect an object. Although we has successfully run these codes, we must understand how YOLO works.

4.1 anchors clustering

The paper suggests to use clustering on bounding box shape to find the good anchor box specialization suited for the data. more details see here

4.2 Architercutre details

In this project, I use the pretrained weights, where we have 80 trained yolo classes (COCO dataset), for recognition. And the class label is represented as c and it's integer from 1 to 80, each number represents the class label accordingly. If c=3, then the classified object is a car. The image features learned by the deep convolutional layers are passed onto a classifier and regressor which makes the detection prediction.(coordinates of the bounding boxes, the class label.. etc).details also see in the below picture. (thanks Levio for your great image!)

4.3 Neural network io:

input : [None, 416, 416, 3]
output : confidece of an object being present in the rectangle, list of rectangles position and sizes and classes of the objects begin detected. Each bounding box is represented by 6 numbers (Rx, Ry, Rh, Rw, Pc, C1..Cn) as explained above. In this case n=80, which means we have c as 80-dimensional vector, and the final size of representing the bounding box is 85. why is 85? see also in the below picture As you can see in the above picture, The first number Pc is the confidence of an project, The second four number bx, by, bh, bw represents the information of bounding boxes. The last 80 number each is the output probability of corresponding-index class.

4.4 Filtering with score threshold

The output result may contain several rectangles that are false positives or overlap, if your input image size of [416, 416, 3], you will get (52X52+26X26+13X13)x3=10647 boxes since YOLO v3 totally uses 9 anchor boxes. (Three for each scale). So It is time to find a way to reduce them. The first attempt to reduce these rectangles is to filter them by score threshold.

Input arguments:

boxes: tensor of shape [10647, 4)]
scores: tensor of shape [10647, 80] containing the detection scores for 80 classes.
score_thresh: float value , then get rid of whose boxes with low score

# Step 1: Create a filtering mask based on "box_class_scores" by using "threshold".
score_thresh=0.4
mask = tf.greater_equal(scores, tf.constant(score_thresh))

4.5 Do non-maximum suppression

Even after yolo filtering by thresholding over, we still have a lot of overlapping boxes. Second approach and filtering is Non-Maximum suppression algorithm.

Discard all boxes with Pc <= 0.4
While there are any remaining boxes :
- Pick the box with the largest Pc
- Output that as a prediction
- Discard any remaining boxes with IOU>=0.5 with the box output in the previous step

In tensorflow, we can simply implement non maximum suppression algorithm like this. more details see here

for i in range(num_classes):
    tf.image.non_max_suppression(boxes, score[:,i], iou_threshold=0.5)

Non-max suppression uses the very important function called "Intersection over Union", or IoU. Here is an exmaple of non maximum suppression algorithm: on input the aglorithm receive 4 overlapping bounding boxes, and the output returns only one