search

zhou13 / neurvps

Neural Vanishing Point Scanning via Conic Convolution
MIT License
176 stars 23 forks source link

how to test your pretrained model on a picture from my own computer #7

Closed ZeyingXuHuaWei closed 1 year ago

ZeyingXuHuaWei commented 4 years ago

hello， I am poor at programming, can you tell me how to test your pretrained model on a picture from my own computer?

zhou13 commented 4 years ago

Currently you need to write you own dataset to test it. I might provide a script for this in the future.

ZeyingXuHuaWei commented 4 years ago

@zhou13 thank you for your reply! May I ask another question. Suppose that three vanishing points of a picture in manhattan world are detected, can we compute the camera pose (including position and rotation ) or can we just obtain camera rotation?

zhou13 commented 4 years ago

Just camera rotation.

hashimK commented 4 years ago

@zhou13 I tried to make my own dataset file by reverse-engineering the datasets.py file and realized that label files like .mat, .txt, .npz are needed for each image to test. But I really want to test any random image with no label data. Hence, request you to provide a script/example that shows how to get vanishing points of any image without having any label data. Thank you.

zhou13 commented 4 years ago

@hashimK The easiest way to do that for now is to create a dataset with a random label. Then you will be able to test it and find the vanishing point without gt labels. To finding the dominant vanishing point, I would suggest use the pre-trained model on Natural Scene dataset.

hashimK commented 4 years ago

@zhou13 Thanks! It helps this way

xiaoxiaoyi689 commented 1 year ago

Hi Zhou! Could you provide the test script?

zhou13 commented 1 year ago

Sorry, I still don't have the scripts. I will close this issue as it is unlikely for me to provide such a script in the near future. However, I will be happy to test one if somebody makes an PR.