TASK-16: Gather Training data for the Waste Recognition Model

[cristian-aldea]

Task Description

As a developer, I want to improve the object detection model by gathering data on objects it has trouble detecting so that it can be a useful tool.

Acceptance Criteria

• [x] Repository of training data is available: link (password on Slack)
• [ ] Data is saved in the Pascal VOC format
• [ ] Gather as much data as possible for the following objects in the Pascal format (list to be expanded as we go):
  • [x] Bottles
  • [x] Food
  • [x] Tin cans
  • [x] Box
  • [ ] ...
  • [ ] After gathering data, some form of accuracy report should be done to assess the performance of the model, and where improvements are needed

Hints:

• Converter which can convert between many standard formats: https://roboflow.com/formats/pascal-voc-xml
  • In general, many tools exist which can convert between different format (ex: "convert to Pascal VOC" on Google). If we can't find what we need, we can just write scripts ourselves and add it to "ml/trash-recognition".
• Good data source: https://storage.googleapis.com/openimages/web/index.html
• Good tool to download from open images: https://pypi.org/project/openimages/
• Good tool to label data manually in the Pascal VOC format: https://github.com/tzutalin/labelImg
• Or maybe even have other models annotate images for us 👀 https://scale.com/image
• ~500 photos minimum per object

[bit172]

Data was gathered. This ticket will be re-opened in the future to refine our model.