Object Detection : printed label detection

---

This project is for detecting the mis-printed label

The model detecte the printed label in real time

• using cv2
• detected printed label using videocam
• finetune the pretrained objecte detection model resnet50

Sample images and Results

printed label

Blurred image

Empty Label

Not detected Label

False Negative Label

Class Activation Heatmap

For production false positive is morre important than false negative

• This models shows the very low false positive rate

For training

• There are few negative(False) example avaialble
• To make up this I genrated Synthetic False exmaples

Generated False Example

Model : 2 step:

• 1. Object detection : Detect the printed label area using

  finetune model : ssd_resnet50_v1_fpn_640x640_coco17

• 2. Classification : Classify the true / false labeled print

  finetune model : mobilenet

requirements

• 1. install required package pip install -r requirements.txt
• 2. install imageai package pip install imageai --upgrade