kwea123/CasMVSNet_pl

Fusion results for all scans using ``` python eval.py \ --dataset_name blendedmvs \ --root_dir /home/ubuntu/data/BlendedMVS/dataset_full_res/ \ --split all \ --ckpt_path ckpts/exp_g8_blended/epoch.15.ckpt \ --num_groups 8 --depth_interval 192.0 ``` and other default parameters in `eval.py`. The point cloud sizes vary from 1M points to maximum 250M points for large scenes. Large scenes require ~10G RAM to open, so pay attention to free your memory before open the files otherwise your pc will freeze. Due to the large size of some scenes, I compressed them. Please decompress before visualization. Put under `results/blendedmvs/points`. Use `python visualize_ply.py --dataset_name blendedmvs --scan {scan name}` to visualize. Take a look at [BlendedMVS_scenes](https://github.com/kwea123/BlendedMVS_scenes) to quickly find out how the scenes look like. Note: scenes `5bbb6eb2ea1cfa39f1af7e0c` and `5b558a928bbfb62204e77ba2` are still more than 2GB after compression, so I cannot put them here. All other 111 scenes are available.

Fusion results for all scans using default parameters in `eval.py` (except that indoor scenes have `--min_geo_consistent=3`). Each point cloud contains 100M~300M points. Due to the large size of some scenes, I compressed them. Please decompress before visualization. Put under `results/tanks/points`. Use `python visualize_ply.py --dataset_name tanks --scan {scan name}` to visualize.

Fusion results for all scans (train, val and test) using default parameters in `eval.py`. Each point cloud contains 20M~30M points. Put under `results/dtu/points`. Use `python visualize_ply.py --dataset_name dtu --scan {scan name}` to visualize. A viewpoint `viewpoint.json` is also provided. add `--use_viewpoint` to use the same viewpoint across scans.

release DTU pretrained model and training logs