Our task-specific networks were trained on 3 million images of varied indoor scenes. Each network was trained on the same distribution of input images. Here, we provide a demo where you can upload (or use a preselected image) and visualize the outputs of different networks. The pretrained models can be downloaded here.
As a rate-limiting method, we ask that you please reload the page (to reload the reCAPTCHA) between successive image uploads. Processing usually takes around 10 seconds.