Can I ask what is the setting value of near/far for that you used?
Posted by: binhnk @ June 5, 2023, 7:18 a.m.Hi binhnk,
Thank you for your question. We have shared the LLFF format (poses_bounds_train.npy) in the dataset alongside the blender format (transforms_train.json). While the blender format only contains “file_path” and “transform_matrix” with “camera_angle_x”, the LLFF format contains additional values such as near (0.8) and far (5.0) values, image resolution (width: 3000, height: 4096), and focal length (15272.744115576517). Please refer to the code of our baseline method (TensoRF) on Github to read the LLFF format: https://github.com/apchenstu/TensoRF/blob/main/dataLoader/llff.py#L151-L165
Although we released our dataset with near and far values of 0.8 and 5.0 respectively, we tested and used tighter bounds of 0.4 and 2.8 to produce the baseline score on the leaderboard. However, there are no restrictions on the values you choose for near and far or any other restrictions in this challenge.
Posted by: youngkyoonjang @ June 5, 2023, 9:50 a.m.Hi youngkyoonjang,
Could you share more information about how you train the TensoRF baseline? I tried following your instructions, but it doesn't work as expected. It'd be great if you could provide more information so that challenge participants could also establish the same baseline on their end.
Posted by: aarush98 @ June 8, 2023, 5:14 p.m.Hi aarush98,
I have created a separate topic to answer your question. Please refer to the topic titled ‘Could you share more information about how you train the TensoRF baseline?’
Posted by: youngkyoonjang @ June 12, 2023, 10:47 a.m.