bmild

Simply change the following flags to the render script to get a smoother video: ``` --gin_bindings="Config.render_path_frames = 480" \ --gin_bindings="Config.render_video_fps = 60" \ ```

You should be able to run something like ``` python -m train \ --gin_configs=configs/360.gin \ --gin_bindings="Config.data_dir = 'my_dataset_dir'" \ --gin_bindings="Config.checkpoint_dir = 'my_dataset_dir/checkpoints'" \ --logtostderr ```

Hi, so far we don't support stitched 360 data but you can use fisheye data (the direct output from one half of the 360 camera) if you use the OPENCV_FISHEYE...

You'll have to change the code that generates the new view rendering path, since the current options all assume roughly planar capture. I wrote some pointers on the relevant code...

Is the flickering at the edges of the window? If so it's probably because the set of MPIs being used to render the new view switches as the new view...

Hmm it looks like there are a couple potential errors... Tensorflow failed to load based on the error `ImportError: libcublas.so.10.0: cannot open shared object file: No such file or directory`...

That's strange. Could be a permissions issue. Maybe try just running the steps in the script one by one on the command line: ``` cd checkpoints wget http://people.eecs.berkeley.edu/~bmild/llff/data/llff_trained_model.zip unzip llff_trained_model.zip...

Sorry I got confused, the poses_bounds.npy file is not included in the download -- it should be generated by the demo script code. Is the error message still the same...

Ah that's just bad error checking on my part, looks like `imgs2mpis.py` was terminated early because of `mkdir: cannot create directory ‘data/testscene/outputs/’: File exists`, then the cuda renderer tried to...

Hmm that is strange, no useful error message. I just realized you are using a smaller GPU, how much GPU memory is there? You may have to mess with the...