As a newbie in computer vision, you can easily be lost in the tooling environment, because there are so many tools available and probably you do not know where to start. We want to have everything as easy as possible and therefore need tools that are handy and easy to use.
For that reason, we show you how you can create fast your first object detection dataset within minutes.
- Gather images (Manthano / Roboflow / Crawler)
- Preselection (WhatToLabel / by hand)
- Annotate Images (Manthano)
- Output Format (Roboflow)
First of all, you need to gather some images, therefore you can either take your own images or use an existing dataset. Within our app Manthano, you can gather them and annotate them in the same tool, but there are also some other existing datasets that you can use. Otherwise, you can also use an image crawler and get some images from google images. There are several available, here is one to use.
Secondly, you can make a preselection of your dataset by removing not appropriate data. There are some tools that make it easy to do so. We recommend WhatToLabel.
Third, you need to make some annotations on your dataset. For the purpose of convenience, we recommend using our annotation tool ManthanoAI. It is super easy to use and you can do your annotations from anywhere.
Last but not least you need the appropriate format for your dataset, roboflow.ai provides a platform that makes it super easy to create your needed data format, as well as tfrecords which can save you a ton of time, believe me ;). Also, you can use their image transformers which makes it super easy to create additional data.
These tools are all freely available for a small number of images and fit perfectly for your first own dataset. Try it out and give us your feedback.