Load a .png or .jpg image from your file system, and discover the amazing predictions!
The model used here has been trained on Mapillary dataset during 5 epochs with a set of 18000 training images. It has been validated with a smaller set of 2000 images.
The model expect street-scene images. You can try whatever picture you want, however in the case you pick another type of scene, do not be surprised if the result looks weird...
The neural network was trained with 400*400 pixel images, that sounds quite small in regard with current camera resolutions. You should expect a more or less important pixelization phenomenon depending on your image size.
As a recall, one can list the 13 Mapillary aggregated labels that were used during the model training:
The uploaded images are stored on our server when the uploading action is done. We do not use them in any other purpose than this web application, and we remove them from the server every two weeks.
The "try-it-yourself" usage implies that the user agrees with the terms associated to the transmitted images.