Multi-GPU training can only be enabled after densification(Try 2.16. New Multiple GPU training strategy)
- Multi-GPU/Node training
- Switch between diff-gaussian-rasterization and nerfstudio-project/gsplat
- Multiple dataset types support
- Interactive web viewer
- Load multiple models
- Model transform
- Scene editor
- Video camera path editor
- Video renderer
- Load a large number of images without OOM
- Dynamic object mask
- Derived algorithms
- Deformable Gaussians
- Deformable 3D Gaussians (2.5.)
- 4D Gaussian (4.3.) (Viewer Only)
- Mip-Splatting (2.6.)
- LightGaussian (2.7.)
- AbsGS / EfficientGS (2.8.)
- 2D Gaussian Splatting (2.9.)
- Segment Any 3D Gaussians (2.10.)
- Large-scale scene reconstruction with partitioning and LoD (2.11.)
- New Appearance Model (2.12.): improve the quality when images have various appearances
- 3D Gaussian Splatting as Markov Chain Monte Carlo (2.13.)
- Feature distillation (2.14.)
- In the wild (2.15.)
- New Multiple GPU training strategy (2.16.)
- SpotLessSplats (2.17.)
- Depth Regularization with Depth Anything V2 (2.18.)
- StopThePop (2.19.)
- Scale Regularization (2.20.)
- Taming 3DGS (2.21.)
- Deformable Gaussians
# clone repository
git clone https://github.com/yzslab/gaussian-splatting-lightning.git
cd gaussian-splatting-lightning
# create virtual environment
conda create -yn gspl python=3.9 pip
conda activate gspl
-
Tested on
PyTorch==2.0.1
-
You must install the one match to the version of your nvcc (nvcc --version)
-
For CUDA 11.8
pip install torch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt
-
ffmpeg is required if you want to render video:
sudo apt install -y ffmpeg
-
If you want to use nerfstudio-project/gsplat
NOTE: Only my modified v1 is supported
pip install git+https://github.com/yzslab/gsplat.git@2646b8ecbd4be78bbc88c2f33f314937755791b8
-
If you need SegAnyGaussian
-
gsplat (see command above)
-
pip install hdbscan scikit-learn==1.3.2 git+https://github.com/facebookresearch/segment-anything.git
-
For
torch==2.0.1
and cuda 11.8:pip install fvcore iopath pip install --no-index --no-cache-dir pytorch3d -f https://dl.fbaipublicfiles.com/pytorch3d/packaging/wheels/py39_cu118_pyt201/download.html
-
Download ViT-H SAM model, place it to the root dir of this repo.:
wget -O sam_vit_h_4b8939.pth https://dl.fbaipublicfiles.com/segment_anything/sam_vit_h_4b8939.pth
-
python main.py fit \
--data.path DATASET_PATH \
-n EXPERIMENT_NAME
It can detect some dataset type automatically. You can also specify type with option --data.parser
. Possible values are: Colmap
, Blender
, NSVF
, Nerfies
, MatrixCity
, PhotoTourism
, SegAnyColmap
, Feature3DGSColmap
.
[NOTE] By default, only checkpoint files will be produced on training end. If you need ply file in vanilla 3DGS's format (can be loaded by SIBR_viewer or some WebGL/GPU based viewer):
- [Option 1]: Convert checkpoint file to ply:
python utils/ckpt2ply.py TRAINING_OUTPUT_PATH
, e.g.:python utils/ckpt2ply.py outputs/lego
python utils/ckpt2ply.py outputs/lego/checkpoints/epoch=300-step=30000.ckpt
- [Option 2]: Start training with option:
--model.save_ply true
- Run training with web viewer
python main.py fit \
--viewer \
...
- It is recommended to use config file
configs/blender.yaml
when training on blender dataset.
python main.py fit \
--config configs/blender.yaml \
...
- With mask (colmap dataset only)
- You may need to undistort mask images too: utils/colmap_undistort_mask.py
# the requirements of mask
# * must be single channel
# * zero(black) represent the masked pixel (won't be used to supervise learning)
# * the filename of the mask file must be image filename + '.png',
# e.g.: the mask of '001.jpg' is '001.jpg.png'
... fit \
--data.parser Colmap \
--data.parser.mask_dir MASK_DIR_PATH \
...
- Use downsampled images (colmap dataset only)
You can use utils/image_downsample.py
to downsample your images, e.g. 4x downsample: python utils/image_downsample.py PATH_TO_DIRECTORY_THAT_STORE_IMAGES --factor 4
# it will load images from `images_4` directory
... fit \
--data.parser Colmap \
--data.parser.down_sample_factor 4 \
...
Rounding mode is specified by --data.parser.down_sample_rounding_mode
. Available values are floor
, round
, round_half_up
, ceil
. Default is round
.
-
Load large dataset without OOM
- [1st option] Cache images in uint8 data type
... fit \ --data.image_uint8 true
- [2nd option] Limit the maximum number of the cached images
- Cache the next batch during training (Recommended)
... fit \ --data.train_max_num_images_to_cache 512 \ --data.async_caching true \ ...
- Cache the next batch at the end of the current batch
... fit \ --data.train_max_num_images_to_cache 1024 \ ...
- Cache the next batch during training (Recommended)
- [1st option] Cache images in uint8 data type
-
Speedup training
-
Store all images in GPU memory
... fit \ --data.image_on_cpu false \ ...
Combining this with
--data.image_uint8 true
to reduce GPU memory consumption is also feasible. -
Avoid performing a validation after every training epoch
Simply set
check_val_every_n_epoch
to a very large value. Please note that the metrics starting withval/
, such asval/psnr
, on the progress bar will not be updated during the training.... fit \ --trainer.check_val_every_n_epoch 99999 \ ...
-
Take a look at Taming 3DGS (2.21.) for further acceleration
-
2.3. Use nerfstudio-project/gsplat
Make sure that command which nvcc
can produce output, or gsplat will be disabled automatically.
python main.py fit \
--config configs/gsplat.yaml \
...
[NOTE] Try New Multiple GPU training strategy, which can be enabled during densification.
[NOTE] Multi-GPU training with DDP strategy can only be enabled after densification. You can start a single GPU training at the beginning, and save a checkpoint after densification finishing. Then resume from this checkpoint and enable multi-GPU training.
You will get improved PSNR and SSIM with more GPUs:
# Single GPU at the beginning
python main.py fit \
--config ... \
--data.path DATASET_PATH \
--model.density.densify_until_iter 15000 \
--max_steps 15000
# Then resume, and enable multi-GPU
python main.py fit \
--config ... \
--trainer configs/ddp.yaml \
--data.path DATASET_PATH \
--max_steps 30000 \
--ckpt_path last # find latest checkpoint automatically, or provide a path to checkpoint file
deform-gs-new.mp4
python main.py fit \
--config configs/deformable_blender.yaml \
--data.path ...
2.6. Mip-Splatting
Training:
python main.py fit \
--config configs/mip_splatting_gsplat_v2.yaml \
--data.path ...
Fuse the 3D smoothing filter to the Gaussian parameters:
python utils/fuse_mip_filter.py \
TRAINED_MODEL_DIR
2.7. LightGaussian
-
Prune & finetune only currently
-
Train & densify & prune
... fit \ --config configs/light_gaussian/train_densify_prune-gsplat.yaml \ --data.path ...
-
Prune & finetune (make sure to use the same hparams as the input model used)
... fit \ --config configs/light_gaussian/prune_finetune-gsplat.yaml \ --data.path ... \ ... \ --ckpt_path YOUR_CHECKPOINT_PATH
2.8. AbsGS / EfficientGS
... fit \
--config configs/gsplat-absgrad.yaml \
--data.path ...
-
Install
diff-surfel-rasterization
firstpip install git+https://github.com/hbb1/diff-surfel-rasterization.git@e0ed0207b3e0669960cfad70852200a4a5847f61
-
Then start training
... fit \ --config configs/vanilla_2dgs.yaml \ --data.path ...
-
Mesh extraction
Install required libraries first:
pip install open3d==0.18.0 scikit-image==0.24.0 trimesh==4.4.3
-
Bounded
python utils/gs2d_mesh_extraction.py MODEL_OUTPUT_PATH
-
Unbounded
python utils/gs2d_mesh_extraction.py MODEL_OUTPUT_PATH --unbounded true
-
2.10. Segment Any 3D Gaussians
-
First, train a 3DGS scene using gsplat
python main.py fit \ --config configs/gsplat.yaml \ --data.path data/Truck \ -n Truck -v gsplat # trained model will save to `outputs/Truck/gsplat`
-
Then generate SAM masks and their scales
-
Masks
python utils/get_sam_masks.py data/Truck/images
You can specify the path to SAM checkpoint via argument
-c PATH_TO_SAM_CKPT
-
Scales
python utils/get_sam_mask_scales.py outputs/Truck/gsplat
Both the masks and scales will be saved in
data/Truck/semantics
, the structure ofdata/Truck
will like this:├── images # The images of your dataset ├── 000001.jpg ├── 000002.jpg ... ├── semantic # Generated by `get_sam_masks.py` and `get_sam_mask_scales.py` ├── masks ├── 000001.jpg.pt ├── 000002.jpg.pt ... └── scales ├── 000001.jpg.pt ├── 000002.jpg.pt ... ├── sparse # colmap sparse database ...
-
-
Train SegAnyGS
python seganygs.py fit \ --config configs/segany_splatting.yaml \ --data.path data/Truck \ --model.initialize_from outputs/Truck/gsplat \ -n Truck -v seganygs # save to `outputs/Truck/seganygs`
The value of
--model.initialize_from
is the path to the trained 3DGS model -
Start the web viewer to perform segmentation or cluster
python viewer.py outputs/Truck/seganygs
SegAnyGS-WebViewer.mp4
Baseline | Partitioning |
---|---|
The implementation here references Block-NeRF, VastGaussian and CityGaussians.
There is no single script to finish the whole pipeline. Please refer to below contents about how to reconstruct a large scale scene.
-
Partitioning
- MatrixCity: notebooks/matrix_city_split.ipynb (Refer to MatrixCity.md about preparing MatrixCity dataset)
- Colmap: notebooks/colmap_split_v2.ipynb
-
Training
[NOTE] You must explicitly specify using a
gsplat
-based renderer (e.g.,configs/gsplat.yaml
,configs/appearance_embedding_renderer
,configs/mip_splatting_gsplat.yaml
, etc) because some parts of the pipeline make assumptions about this. If you do not, it will usediff-gaussian-rasterization
by default, which will cause a drop in the final quality.- MatrixCity: utils/train_matrix_city_partitions_v2.py
- Colmap: utils/train_colmap_partitions_v2.py
-
Optional LightGaussian pruning
- Pruning: utils/prune_partitions_v2.py
- Finetune after pruning: utils/finetune_pruned_partitions_v2.py
-
Merging: utils/merge_partitions_v2.py
-
First prepare the dataset
# 1. download the dataset mkdir -p data/MegaNeRF pushd data/MegaNeRF wget -O rubble-pixsfm.tgz https://storage.cmusatyalab.org/mega-nerf-data/rubble-pixsfm.tgz tar -zxf rubble-pixsfm.tgz popd # 2. create a colmap sparse model from provided camera poses wget -O vocab_tree_flickr100K_words256K.bin https://demuc.de/colmap/vocab_tree_flickr100K_words256K.bin python utils/meganerf2colmap.py data/MegaNeRF/rubble-pixsfm -v vocab_tree_flickr100K_words256K.bin # 3. down sample images python utils/image_downsample.py data/MegaNeRF/rubble-pixsfm/colmap/images --factor 3
The intrinsics and extrinsics provided in the
Rubble
dataset seem not optimal and will produce a slightly blurry result. Run a sparse reconstruction from scratch if you want to avoid it. -
Generate appearance groups
The Rubble dataset contains images in various lighting conditions. Enabling the appearance model can improve the quality. In order to do so, appearance groups must be generated first.
python utils/generate_image_apperance_groups.py \ data/MegaNeRF/rubble-pixsfm/colmap \ --image \ --name appearance_image_dedicated
-
Partitioning: simply use notebooks/meganerf_rubble_split.ipynb, and remember to change the value of
dataset_path
in this notebook -
Training
PARTITION_DATA_PATH="data/MegaNeRF/rubble-pixsfm/colmap/partitions-size_60.0-enlarge_0.1-visibility_0.9_0.25" PROJECT_NAME="MegaNeRF-rubble" python utils/train_colmap_partitions_v2.py \ ${PARTITION_DATA_PATH} \ -p ${PROJECT_NAME} \ --scalable-config utils/scalable_param_configs/appearance.yaml \ --config configs/appearance_embedding_renderer/sh_view_dependent.yaml \ -- \ --data.parser.appearance_groups appearance_image_dedicated \ --model.gaussian.optimization.spatial_lr_scale 15 \ --data.parser.down_sample_factor 3
Parameter Explanations
utils/train_colmap_partitions_v2.py
- The first is the path of the directory containing the partition data, it is the value of
output_path
in the partitioning notebook -p
: the project name, all the trained partition models will be stored inoutputs/PROJECT_NAME
--scalable-config
: the path of the yaml file specifying the hyperparameters that will be adjusted according to the image number--config
: the config file used for training--
: all the parameters after this will be passed tomain.py
as-is
- The first is the path of the directory containing the partition data, it is the value of
main.py
-
--data.parser.appearance_groups
: the name of the generated appearance groups -
--model.gaussian.optimization.spatial_lr_scale
: the LR of the 3D means of Gaussians will be multiplied by this value, which determines how finely the structure can be modeledBy default, this value will be calculated automatically according to the camera poses if it is omitted, but this will lead to suboptimal results for large-scale scenes. Therefore you had better provide it manually.
Please note that this is a scene-specific hyperparameter. You should not use the same value for another scene, or even when you run a colmap sparse reconstruction a second time. Additionally, the value
15
is not the optimal one for the dataset Rubble. -
--data.parser.down_sample_factor
: down sample factor of the images
-
-
Merging
python utils/merge_partitions_v2.py \ ${PARTITION_DATA_PATH} \ -p ${PROJECT_NAME}
Then you can start the web viewer with the merged checkpoint file.
-
Optional prune and finetune
-
Prune
python utils/prune_partitions_v2.py \ ${PARTITION_DATA_PATH} \ -p ${PROJECT_NAME}
-
Finetune
PRUNED_PROJECT_NAME="${PROJECT_NAME}-pruned" python utils/finetune_pruned_partitions_v2.py \ ${PARTITION_DATA_PATH} \ -p ${PRUNED_PROJECT_NAME} \ -t ${PROJECT_NAME} \ --scalable-config utils/scalable_param_configs/appearance.yaml
It will load trained model from
outputs/${PROJECT_NAME}
, and the finetuned outputs will be saved tooutputs/${PRUNED_PROJECT_NAME}
. -
Merge finetuned outputs
python utils/merge_partitions_v2.py \ ${PARTITION_DATA_PATH} \ -p ${PRUNED_PROJECT_NAME}
-
-
LoD Rendering
With LoD, the renderer will select the finer models for partitions close to the camera, and coarser models for those far away.
-
(a) Preprocess partitions' checkpoints
This is done by simply add an option
--preprocess
when runningutils/merge_partitions_v2.py
. You need to run it for every level.# First for the original models python utils/merge_partitions_v2.py \ --preprocess \ ${PARTITION_DATA_PATH} \ -p ${PROJECT_NAME} # Then for the pruned models python utils/merge_partitions_v2.py \ --preprocess \ ${PARTITION_DATA_PATH} \ -p ${PRUNED_PROJECT_NAME}
-
(b) Create a LoD config file in YAML
# `data` is the path to your partition data data: data/MegaNeRF/rubble-pixsfm/colmap/partitions-size_60.0-enlarge_0.1-visibility_0.9_0.25 # `names` is a list of project names, where order represents their detail levels, ranging from fine to coarse names: - MegaNeRF-rubble # without pruning - MegaNeRF-rubble-pruned # pruned # - MegaNeRF-rubble-pruned_again # add more lines if you have
-
(c) Start the viewer
python viewer YOU_LOD_CONFIG_FILE_PATH.yaml
-
-
Train/prune/finetune multiple partitions in parallel
The options
--n-processes
and--process-id
are designed for the parallel purpose.--n-processes
is the total number of GPUs you want to use, which has the same meaning as world size.--process-id
is the unique process id, just like the global rank, but ranging from 1 to the value of--n-processes
here. Please note that both of the options should be placed before--
.Assuming you have 2 machines, and both of them are equipped with 2 GPUs:
-
On the first machine
First run:
CUDA_VISIBLE_DEVICES=0 python utils/train_colmap_partitions_v2.py \ ... \ --n-processes 4 \ --process-id 1 \ -- \ ...
Then run this in another terminal:
CUDA_VISIBLE_DEVICES=1 python ... --n-processes 4 --process-id 2
-
On the second machine
First:
CUDA_VISIBLE_DEVICES=0 python ... --n-processes 4 --process-id 3
Then another terminal:
CUDA_VISIBLE_DEVICES=1 python ... --n-processes 4 --process-id 4
Or submit partition training jobs to Slurm:
# Adding params after the 2nd `--` means submitting jobs to Slurm, # and those params belong to `srun`. # Do not add `--n-processes` and `--process-id` here. python utils/train_colmap_partitions_v2.py \ ... \ -- \ ... \ -- \ --gres=gpu:1 # one gpu per-partition training job
-
-
Train/finetune a partition with multiple GPUs
- Training only works with the strategy introduced in 2.16.
- Finetune only works with DDP
With appearance model, the reconstruction quality can be improved when your images have various appearance, such as different exposure, white balance, contrast and even day and night.
This model assign an extra feature vector
Please refer to internal/renderers/gsplat_appearance_embedding_renderer.py for more details.
Baseline | New Model |
---|---|
Train-head-baseline.mp4 |
Train-head.mp4 |
Day-and-Night.mp4 |
-
First generate appearance groups (Colmap or PhotoTourism dataset only)
python utils/generate_image_apperance_groups.py PATH_TO_DATASET_DIR \ --image \ --name appearance_image_dedicated # the name will be used later
The images in a group will share a common appearance embedding. The command above will assign each image a group, which means that will not share any appearance embedding between images.
-
Then start training
python main.py fit \ --config configs/appearance_embedding_renderer/view_dependent.yaml \ --data.path PATH_TO_DATASET_DIR \ --data.parser Colmap \ --data.parser.appearance_groups appearance_image_dedicated # value here should be the same as the one provided to `--name` above
If you are using PhotoTourism dataset, please replace
--data.parser Colmap
with--data.parser PhotoTourism
. -
view_independent.yaml
: turn off view dependent effectssh_view_dependent.yaml
: represent view dependent effects using spherical harmonics*-distributed.yaml
: multiple GPUs*-estimated_depth_reg.yaml
/*-estimated_depth_reg-hard_depth.yaml
: with depth regularization
-
Remove the dependence on MLP when rendering
It is recommended to use
view_independent-*
orsh_view_dependent-*
configs if you want to do so.By running
python utils/fuse_appearance_embeddings_into_shs_dc.py TRAINED_MODEL_DIR
, you can get a fixed appearance checkpoint without requiring a MLP.
2.13. 3DGS-MCMC
... fit \
--config configs/gsplat-mcmc.yaml \
--model.density.cap_max MAX_NUM_GAUSSIANS \
...
MAX_NUM_GAUSSIANS
is the maximum number of Gaussians that will be used.
Refer to ubc-vision/3dgs-mcmc, internal/density_controllers/mcmc_density_controller.py and internal/metrics/mcmc_metrics.py for more details.
Click me
This comes from Feature 3DGS. But two stage optimization is adapted here, rather than jointly.
-
First, train a model using gsplat (see command above)
-
Then extract feature map from your dataset
Theoretically, any feature is distillable. You need to implement your own feature map extractor. Here are instructions about extracting SAM and LSeg features.
-
SAM
python utils/get_sam_embeddings.py data/Truck/images
With this command, feature maps will be saved to
data/Truck/semantic/sam_features
, and preview todata/Truck/semantic/sam_feature_preview
, respectively. -
LSeg: please use ShijieZhou-UCLA/feature-3dgs and follow its instruction to extra LSeg features (do not use this repo's virtual environment for it).
-
-
Then start distillation
-
SAM
python main.py fit \ --config configs/feature_3dgs/sam-speedup.yaml \ --data.path data/Truck \ --data.parser.down_sample_factor 2 \ --model.initialize_from outputs/Truck/gsplat \ -n Truck -v feature_3dgs-sam
-
LSeg
[NOTE] In order to distill LSeg's high-dimensional features, you may need a GPU equipped with a large memory capacity
python main.py fit \ --config configs/feature_3dgs/lseg-speedup.yaml \ ...
--model.initialize_from
is the path to your trained model.Since rasterizing high dimension features is slow,
--data.parser.down_sample_factor
is used here to smaller the rendered feature map to speedup distillation. -
-
After distillation finishing, you can use viewer to visualize the feature map rendered from 3D Gaussians
python viewer.py outputs/Truck/feature_3dgs
CLIP is required if you are using LSeg feature:
pip install git+https://github.com/openai/CLIP.git
object-recognition.mp4
LSeg feature is used in this video.
Based on the Appearance Model (2.12.) above, this model can produce a visibility map for every training view indicating whether a pixel belongs to transient objects or not.
The idea of the visibility map is a bit like Ha-NeRF, but rather than uses positional encoding for pixel coordinates, 2D dense grid encoding is used here in order to accelerate training.
Please refer to Ha-NeRF, internal/renderers/gsplat_appearance_embedding_visibility_map_renderer.py
and internal/metrics/visibility_map_metrics.py
for more details.
[NOTE] Though it shows the capability to distinguish the pixels of transient objects, may not be able to remove some artifats/floaters belong to transients. And may also treat under-reconstructed regions as transients.
- tiny-cuda-nn is required
pip install git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch
- Preparing dataset
Download PhotoTourism dataset from here and split file from the "Additional links" here. The split file should be placed at the same path as the dense
directory of the PhotoTourism dataset, e.g.:
├──brandenburg_gate
├── dense # colmap database
├── images
├── ...
├── sparse
...
├── brandenburg.tsv # split file
[Optional] 2x downsize the images: python utils/image_downsample.py data/brandenburg_gate/dense/images --factor 2
- Start training
python main.py fit \
--config configs/appearance_embedding_visibility_map_renderer/view_independent-2x_ds.yaml \
--data.path data/brandenburg_gate \
-n brandenburg_gate
If you have not downsized images, remember to add a --data.parser.down_sample_factor 1
to the command above.
- Validation on training set
python main.py validate \
--save_val \
--val_train \
--config outputs/brandenburg_gate/lightning_logs/version_0/config.yaml # you may need to change this path
Then you can find the rendered masks and images in outputs/brandenburg_gate/val
.
This is a bit like a simplified version of Scaling Up 3DGS.
In the implementation here, Gaussians are stored, projected and their colors are calculated in a distributed manner, and each GPU rasterizes a whole image for a different camera. No Pixel-wise Distribution currently.
This strategy works with densification enabled.
[NOTE]
- Not well validated yet, still under development
- Multiple GPUs training only currently
- In order to combine with derived algorithms containing neural networks, you need to manually wrap your networks with DDP, e.g.: internal/renderers/gsplat_distributed_appearance_embedding_renderer.py
Metrics of MipNeRF360 dataset
One batch per GPU, 30K iterations, no other hyperparameters changed.- Training
python main.py fit \
--config configs/distributed.yaml \
...
By default, all processes will hold a (redundant) replica of the dataset in memory, which may cause CPU OOM. You can avoid this by adding the option --data.distributed true
, so that each process loads a different subset of the dataset.
- Merge checkpoints
python utils/merge_distributed_ckpts.py outputs/TRAINED_MODEL_DIR
- Start viewer
python viewer.py outputs/TRAINED_MODEL_DIR/checkpoints/MERGED_CHECKPOINT_FILE
2.17. SpotLessSplats
[NOTE] No utilization-based pruning (4.2.3 of the paper) and appearance modeling (4.2.4 of the paper)
-
Install requirements
pip install diffusers==0.27.2 transformers==4.40.1 scikit-learn
-
Extract Stable Diffusion features
python utils/sd_feature_extraction.py YOUR_IMAGE_DIR
-
Training
- Spatial clustering (SLS-agg, 4.1.1)
python main.py fit \ --config configs/spot_less_splats/gsplat-cluster.yaml \ --data.parser.split_mode "reconstruction" \ --data.path YOUR_DATASET_PATH \ -n EXPERIMENT_NAME
- Spatio-temporal clustering (SLS-mlp, 4.1.2)
python main.py fit \ --config configs/spot_less_splats/gsplat-mlp.yaml \ --data.parser.split_mode "reconstruction" \ --data.path YOUR_DATASET_PATH \ -n EXPERIMENT_NAME
- Other available configs
- gsplat-mlp-with_ssim.yaml: with SSIM metric
- gsplat-mlp-opacity_reg_0.01.yaml: with opacity regularization, aiming to reduce floaters/artifacts
- view_independent-phototourism-sls-opacity_reg_0.01.yaml: with new appearance model (2.12.) (not the one mentioned in the SLS paper)
Change the value of
--data.parser.split_mode
tokeyword
if you are using the RobustNeRF dataset. - Spatial clustering (SLS-agg, 4.1.1)
-
Render SLS predicted masks
python utils/render_sls_masks.py outputs/EXPERIMENT_NAME
2.18. Depth Regularization with Depth Anything V2
This is implemented with reference to Hierarchical 3DGS.
Baseline | DepthReg | DepthReg + AppearanceModel |
---|---|---|
baseline.mp4 |
depth_reg.mp4 |
appearance-depth_reg.mp4 |
-
Setup Depth Anything V2
# clone the repo. git clone https://github.com/DepthAnything/Depth-Anything-V2 utils/Depth-Anything-V2 # NOTE: do not run `pip install -r utils/Depth-Anything-V2/requirements.txt` # download the pretrained model `Depth-Anything-V2-Large` mkdir utils/Depth-Anything-V2/checkpoints wget -O utils/Depth-Anything-V2/checkpoints/depth_anything_v2_vitl.pth "https://huggingface.co/depth-anything/Depth-Anything-V2-Large/resolve/main/depth_anything_v2_vitl.pth?download=true"
-
Dataset pre-processing
python utils/estimate_dataset_depths.py data/Family
Make sure that both the
sparse
andimages
folders exist indata/Family
.With the operation above, the structure of
data/Family
should be like this:├── data/Family ├── estimated_depths # generated by `utils/run_depth_anything_v2.py` ├── 00001.jpg.npy ├── ... ├── images ├── 00001.jpg ├── ... ├── sparse # colmap sparse model ├── ... ├── estimated_depth_scales.json # generated by `utils/get_depth_scales.py` ...
-
Training
python main.py fit \ --config configs/depth_regularization/estimated_inverse_depth-l1.yaml \ --data.path data/Family \ -n EXPERIMENT_NAME
Other available configs:
- estimated_inverse_depth-l1_ssim.yaml: with SSIM as an extra depth metric
- estimated_inverse_depth-l2.yaml: L2 depth loss
- estimated_inverse_depth-hard_depth-l1.yaml: better at removing floaters/artifacts
- estimated_inverse_depth-hard_depth-l1_ssim.yaml
- with new appearance model (2.12.)
In my experiments, simply L1 is slightly better than L2 or the one with SSIM.
2.19. StopThePop
-
Install the StopThePop-Rasterization first:
pip install dacite git+https://github.com/yzslab/StopThePop-Rasterization.git
-
Training:
python main.py fit \ --config configs/stp/baseline.yaml \ --data.path ... \ ...
The scales of Gaussians will grow to some unreasonable values after densification. For example, some linear shape Gaussians are almost longer than your scene, and appear as artifacts at many viewpoints. This regularization, containing max scale and scale ratio losses, can avoid it. Take a look internal/metrics/scale_regularization_metrics.py for more details.
Usage:
python main.py fit \
--config configs/scale_reg.yaml \
--model.metric.max_scale 1. \
...
The --model.metric.max_scale
is a scene-specific hyperparameter. The regularization will be applied to the Gaussians with scales exceeding it. It should be greater than percent_dense * camera_extent
. The percent_dense
is 0.01
by default. The camera_extent
will be printed as spatial_lr_scale=...
at the beginning of the training. Set it to a very large value, e.g. 2048
, to disable the max scale loss if you are not sure what value should be used.
2.21. Taming 3DGS
There are two implementations: one is the gsplat v1 based, and the other is the vanilla one. The gsplat v1 based implementation currently does not have "Backpropagation with Per-Splat Parallelization."
-
(a) Install libraries first
-
fused-ssim
pip install git+https://github.com/rahul-goel/fused-ssim.git@d99e3d27513fa3563d98f74fcd40fd429e9e9b0e
-
my modified gsplat v1 if you want the gsplat v1 based one (refer to 1.5. for the setup guide)
-
another rasterizer if you want the vanilla one
pip install git+https://github.com/yzslab/diff-gaussian-rasterization.git@b403ab6c5cfb4ed89265a9759bd4766f9c4b56de
-
-
(b) Available config files
Type gsplat v1 vanilla Competitive quality configs/gsplat_v1-accel.yaml
configs/taming_3dgs/rasterizer-fused_ssim-aa.yaml
More acceleration, slightly lower quality (SparseAdam) configs/gsplat_v1-accel_more.yaml
configs/taming_3dgs/rasterizer-fused_ssim-sparse_adam-aa.yaml
Per-image metrics will be saved to TRAINING_OUTPUT/metrics
as a csv
file.
python main.py validate \
--config outputs/lego/config.yaml
python main.py test \
--config outputs/lego/config.yaml
python main.py validate \
--config outputs/lego/config.yaml \
--val_train
python main.py <validate or test> \
--config outputs/lego/config.yaml \
--save_val
Then you can find the images in outputs/lego/<val or test>
.
Transform | Camera Path | Edit |
---|---|---|
transform.mp4 |
animation.mp4 |
edit.mp4 |
- Also works for graphdeco-inria/gaussian-splatting's ply output
python viewer.py TRAINING_OUTPUT_PATH
# e.g.:
# python viewer.py outputs/lego/
# python viewer.py outputs/lego/checkpoints/epoch=300-step=30000.ckpt
# python viewer.py outputs/lego/baseline/point_cloud/iteration_30000/point_cloud.ply # only works with VanillaRenderer
python viewer.py \
outputs/garden \
outputs/lego \
outputs/Synthetic_NSVF/Palace/point_cloud/iteration_30000/point_cloud.ply \
--enable_transform
[NOTE] The commands in this section only design for third-party outputs
python viewer.py \
Deformable-3D-Gaussians/outputs/lego \
--vanilla_deformable \
--reorient disable # change to enable when loading real world scene
python viewer.py \
4DGaussians/outputs/lego \
--vanilla_gs4d
# Install `diff-surfel-rasterization` first
pip install git+https://github.com/hbb1/diff-surfel-rasterization.git@e0ed0207b3e0669960cfad70852200a4a5847f61
# Then start viewer
python viewer.py \
2d-gaussian-splatting/outputs/Truck \
--vanilla_gs2d
python viewer.py \
SegAnyGAussians/outputs/Truck \
--vanilla_seganygs
python viewer.py \
mip-splatting/outputs/bicycle \
--vanilla_mip
Q: The viewer shows my scene in unexpected orientation, how to rotate the camera, like the U
and O
key in the SIBR_viewer?
A: Check the Orientation Control
on the right panel, rotate the camera frustum in the scene to the orientation you want, then click Apply Up Direction
.
reorient-camera-up.mp4
Besides: You can also click the 'Reset up direction' button. Then the viewer will use your current orientation as the reference.
- First use mouse to rotate your camera to the orientation you want
- Then click the 'Reset up direction' button
Q: The web viewer is slow (or low fps, far from real-time).
A: This is expected because of the overhead of the image transfer over network. You can get around 10fps in 1080P resolution, which is enough for you to view the reconstruction quality.