Dataset Merge guidance

🤝 Community Help
1

So, I am working on a model which detect road anomaly, I have 2 different datasets
First dataset has classes such as HMV, LMV, RoadDamages, UnsurfacedRoads and so on
Second Dataset has class such as Alligator cracking, longitudinal cracking, Potholes and so on

My query is I want to classify the first Dataset into the more class like expand on the RoadDamages as similar to Second Dataset and use that dataset to train my model. Any suggestion how i can do it. Would be really helpful as I am on a time constraint. Thanks

2

Hi @maddox095 ,

classify the first Dataset into the more class like expand on the RoadDamages

Could you elaborate on this? First dataset has classification task, and the second uses classes (detection? segmentation?)? Could you share a link to these datasets, so we can provide the best suggestions?

Thanks, Erik

3

Dataset 1: RDD 2022 | Kaggle
Dataset 2: RAD(Road Anomaly Detection) | Kaggle
It is detection based. I have linked the dataset above

4

Hi @maddox095 , seems like both of these datasets are object detections. You can download them from Kaggle and upload them to Roboflow via CLI or python sdk. And once dataset is on the roboflow you can train a custom model on it.

5

@erik_roboflow yeah, but the training on both dataset causes the model to confuse, cause RoadDamages is the same object as All the classes in Dataset 2
For example potholes (in Dataset 2) is consider RoadDamages (in Dataset 1) so how would the model understand that it has to differentiate RoadDamages. I hope Im clarifying my problem.

6

Hi @maddox095 , I see 2 options:

  • Remove that class/images from the Dataset 1
  • Update classes of Dataset 1 to match the ones from Dataset 2

Which one to choose really depends on your time and how large your dataset is (just removing some classes won’t matter as much if you have huge dataset)

Thoughts?