We used a subset of synthetic images that only contained forward-looking images from the original Unimelb corridor dataset, and removed the additional images that were generated by rotating the camera along the X and Y axes. To compensate for the low number of synthetic images, we generated 900 more images along the original trajectory by reducing the spacing between the consecutive images, which finally resulted in 1400 images for the synthetic dataset. The dataset also contains 950 real images and their corresponding groundtruth camera poses in the BIM coordinate system. We removed some of the redundant images (100) at the end of the trajectory and added another 500 new real images, which resulted in 1350 real images. The synthetic and real cameras have identical intrinsic camera parameters, with an image resolution of 640 x 480 pixels.
Additionally, the provided Blender files can be used to render the images. Please note that SynCar dataset should be rendered with Blender 2.78 only, whereas SynPhoReal and SynPhoRealTex images can be generated using the latest Blender 3.4.
 Acharya, D., Khoshelham, K. and Winter, S., 2019. BIM-PoseNet: Indoor camera localisation using a 3D indoor model and deep learning from synthetic images. ISPRS Journal of Photogrammetry and Remote Sensing, 150, pp.245-258.
 Acharya, D., Singha Roy, S., Khoshelham, K. and Winter, S., 2020. A recurrent deep network for estimating the pose of real indoor images from synthetic image sequences. Sensors, 20(19), p.5492.
 Acharya, D., Tennakoon, R., Muthu, S., Khoshelham, K., Hoseinnezhad, R. and Bab-Hadiashar, A., 2022. Single-image localisation using 3D models: Combining hierarchical edge maps and semantic segmentation for domain adaptation. Automation in Construction, 136, p.104152.
User Contributed Tags
Login to tag this record with meaningful keywords to make it easier to discover