Vase and Table (Better)

1. I got the dataset from avsupport. He gave the raw video with dual fisheye camera. Recorded by Qoocam360 Ultra 2. extracted the video 1 frame per second with ffmpeg. I got 96 images. 3. split the images into top and bottom fisheye. Totalling 192 images, moved that file in the different folder 4. colmap feature_extractor, colmap vocab_tree_matcher, colmap mapper, colmap image_undistorter, colmap patch_match_stereo, and colmap stereo_fusion. I used `--ImageReader.single_camera_per_folder 1 --ImageReader.camera_model THIN_PRISM_FISHEYE`. It runs for 3 hours with RTX4090 in the cloud GPU. After distorting the image, the resolution of each image is 7680x7680 5. I use splatfacto-mcmc. The settings are 30K steps, 2 M splats, downscale-factor=4 (since the original dataset way to huge for VRAM)