Data labeling Specification

0. The data that we use for training will be in form of image.
1. We need to collect glass, Aluminum, and plastic.
2. We will label the objects with color.
3. We need to label objects based on type. Note: plastics have 7 types: PET, HDPE, PVC, LDPE, PP, PS, and Others.
4. What is the perspective of the camera? Images are better to be taken from top.
5. We need to regulate brightness and contrast.
6. We need balanced and diverse datasets. Due to expand the data set we need to use augmentation methods.
7. We need deformed and dirty objects in our dataset.
8. Dark background is better.
9. Due to detecting an object in the image we have to set bounding box for every object, omit the background and detecting the edges.
10. The images should have same formats. But they will convert to another formats when it feed into the detection method.

---

We will have two different kinds of labeling:

• Manual
• Automatic

Here we use Label Studio (https://labelstud.io/) as a tool for labeling images.