robbyant/lingbot-map — reverse-engineered prompt

Build me a Python project that can take a folder of image frames from a moving camera and reconstruct the scene as a 3D map while the frames stream in. I want it to use a pretrained model checkpoint, let me point to the model file and an image folder from the command line, then open a local browser viewer where I can inspect the camera path and the reconstructed scene.

Please include a simple demo command with bundled example scenes, plus options for long sequences, keyframe spacing, sky masking for outdoor scenes, and a lower memory mode if the GPU is struggling. I also want an offline rendering script for very long videos so I can process thousands of frames and save a nice rendered output without using the interactive viewer.

Set up the project so installation is straightforward, with clear README instructions for the environment, model download, running the demo, and running benchmarks if datasets are prepared. Look up current docs online if needed.