sarapohland/speed-limit-signs

This dataset was obtained from the German Traffic Sign Detection Benchmark (GTSDB). The dataset contains 12,570 RGB images (sized 200x200) of common speed limit signs in Germany. All images are labeled as corresponding to one of seven speed limits:

1. 1.30 km/hr (2,220 images)
2. 2.50 km/hr (2,250 images)
3. 3.60 km/hr (1,410 images)
4. 4.70 km/hr (1,980 images)
5. 5.80 km/hr (1,860 images)
6. 6.100 km/hr (1,440 images)
7. 7.120 km/hr (1,410 images)

This labeled dataset has been randomly partitioned into a training set of 9,050 images, a validation set of 2,263 images, and a test set of 1,257 images.

This dataset also contains a set of out-of-distribution (OOD) data points. The in-distribution data contains images only of speed limit signs commonly oberserved in Germany (30 km/hr and higher). OOD data contains images of an uncommon speed limit sign (20 km/hr). There are 210 RGB images (sized 200x200) in the OOD set.

The compressed NumPy file contains the following elements: traindata, trainlabels, valdata, vallabels, testdata, testlabels, ooddata, and oodlabels. Sample code for interacting with this dataset is available in our GitHub repository. In particular, src/datasets/custom_dataset.py demonstrates how to set up a PyTorch Dataset and src/datasets/setup_dataloader.py demonstrates how to set up a PyTorch Dataloader.

Each pixel in each of the OOD images is either familiar or unfamiliar to the perception model, depending on whether it corresponds to structures present in the training set or not. These familiarity labels are stored in the oodlabels pickle file. To adjust these labels, you can use the `src/datasets/createlabels.py` script available in our GitHub repository.