Street View Signboard Objects (SVSO)

Street View Signboard Objects (SVSO), a dataset that contains a total of 5000 signboard images of establishments. All images are in PNG file format with bit depth of 32 bits per pixel. The SVSO data includes signboard images of various shapes and colors from Dhaka, Chottogram, Sylhet, Mymensingh, Barisal, Rajshahi, Khulna, Rangpur, Bogra and Pabna, 10 different cities of Bangladesh. There is also an independent test set containing 200 signboard images from 6 developing countries such as Bangladesh, Bhutan, Malaysia, Myanmar, Sri Lanka and Thailand. In this collection, every image file has been downscaled to 1000*600 pixels using the Anti-aliasing method. Then image normalization has been performed on all images to transform each pixel value between 0 and 1. To keep all image features on the same scale, standardization method has been applied. SVSO was originally created for the research work of Signboard Detection in densely populated areas which is currently the largest existing dataset for signboard class objects.