Kevin Chen

When there are lots of training images, bad things happen

2

**Is your feature request related to a problem? Please describe.** If I am loading lots of images (e.g. ~5000) with the `Nerfstudio` data parser, two things happen: 1. I have...

This PR changes the default value of `--pipeline.datamanager.train-num-images-to-sample-from` to 300 to avoid OOM when caching train data for large datasets. Also, the configuration `--pipeline.datamanager.train-num-times-to-repeat-images` is added which controls how often...

This fixes `activation` being applied on the last layer: ``` ... -> linear -> `activation` -> `output_activation` -> (output tensor) ``` Also replacing tuple-lookup with set-lookup.

Add Support for Equirectangular Images

2

This PR adds support for equirectangular images. The ray directions should be correctly created based on the image type, but the pixel sampler needs to be manually set by a...