The panoramic view is not stiched well (first and last image are looking up or down with non-zero elevation)!

[lukewenMX]

The generated panoramic view from the skybox images is not stiched well by runing the downsize_skybox.py, that is the _skybox_small.jpg is not an accurate panoramic view especially for the first and last images (one is looking up and one is looking down with an non-zero elevation) . An instance has been shown in below, please give some comments and suggestions, thanks.

[peteanderson80]

This is working as intended. The image is a skybox / cubemap not an equirectangular panorama, so it's actually 6 different images representing the faces of a cube and they are just joined together for convenience (it speeds up file loading to have fewer and larger files).

[lukewenMX]

@peteanderson80 Hi, Peter. Thanks for your sharp reply and the answer really helps me. By the way, do you have any ideas about how to generate the 'equirectangular panorama' shown as perfectly stiched together? I did not find any program which supports to load the well-trained agent to interact with Matterport 3D simulator and at the meantime showing the "equirectangular panorama" . Will appreciate your further answers and any suggestions. Thank again.

[lukewenMX]

@peteanderson80 Hi, Peter. I think I just got the idea of the skybox, and the question about the equirectangular panorama，please just ignore. Many thanks.

[lukewenMX]

By the way, is there any convenient way to show the traversed path by the agent from the top-down view of the simulated house, as dipicted below (Picture from the published paper reference: Li, Xiujun, Chunyuan Li, Qiaolin Xia, Yonatan Bisk, Asli Celikyilmaz, Jianfeng Gao, Noah Smith, and Yejin Choi. "Robust navigation with language pretraining and stochastic sampling." arXiv preprint arXiv:1909.02244 (2019).): Many thanks in advance for any help and suggestions.

[jdiazram]

Hi @lukewenMX, could you solve the creation of an equirectangular panorama from skyboxes? I used py360convert, but to analyze the result, py360convert is based on:

Then, after analyzing the original paper, "Vision-and-Language Navigation: Interpreting visually-grounded navigation instructions in real environments," on page 11, I think the heading is another. Like this: