### Test Configuration Options

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Example configuration settings for testing, including saving various outputs like images, depths, and reconstructed Gaussians.

```text
dataset.test_chunk_interval=1
output_dir=outputs/depthsplat
test.save_image=true
test.save_gt_image=true
test.save_input_images=true
test.save_depth=true
test.save_depth_concat_img=true
test.save_depth_npy=true
test.save_gaussian=true
```

--------------------------------

### Install PyTorch and Dependencies

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Installs PyTorch with CUDA support and other project requirements. Ensure you have a compatible CUDA version installed.

```bash
pip install torch==2.4.0 torchvision==0.19.0 --index-url https://download.pytorch.org/whl/cu124
pip install -r requirements.txt
```

--------------------------------

### Render Video on re10k (512x960)

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Renders a video on the re10k dataset with specified configurations for image shape, model encoder, and checkpoint. Ensure ffmpeg is installed.

```bash
CUDA_VISIBLE_DEVICES=0 python -m src.main +experiment=dl3dv \
dataset.test_chunk_interval=1 \
dataset.roots=[datasets/re10k_720p] \
dataset.image_shape=[512,960] \
dataset.ori_image_shape=[720,1280] \
model.encoder.num_scales=2 \
model.encoder.upsample_factor=4 \
model.encoder.lowest_feature_resolution=8 \
model.encoder.monodepth_vit_type=vitb \
model.encoder.gaussian_adapter.gaussian_scale_max=0.1 \
checkpointing.pretrained_model=pretrained/depthsplat-gs-base-re10kdl3dv-448x768-randview2-6-f8ddd845.pth \
mode=test \
dataset/view_sampler=evaluation \
dataset.view_sampler.num_context_views=6 \
dataset.view_sampler.index_path=assets/re10k_ctx_6v_video.json \
test.save_video=true \
test.compute_scores=false \
test.render_chunk_size=10 \
output_dir=outputs/depthsplat-re10k-512x960
```

--------------------------------

### Download Depth Anything V2 Pre-trained Weights

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Downloads the Depth Anything V2 pre-trained weights for model training. Save to the 'pretrained' directory.

```bash
wget https://huggingface.co/depth-anything/Depth-Anything-V2-Small/resolve/main/depth_anything_v2_vits.pth -P pretrained
```

--------------------------------

### Create and Activate Venv Environment

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Sets up a Python 3.10 environment using venv for the project. This helps manage dependencies.

```bash
# or venv
# python -m venv /path/to/venv/depthsplat
# source /path/to/venv/depthsplat/bin/activate
```

--------------------------------

### DepthSplat Zero-Shot DL3DV Test (2 Context Views)

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Command for zero-shot generalization from RealEstate10K to DL3DV, using 2 context views. Ensure pre-trained weights are available.

```bash
# Table 8 of depthsplat paper
CUDA_VISIBLE_DEVICES=0 python -m src.main +experiment=dl3dv \
mode=test \
dataset/view_sampler=evaluation \
dataset.view_sampler.num_context_views=2 \
dataset.view_sampler.index_path=assets/dl3dv_start_0_distance_10_ctx_2v_tgt_4v.json \
model.encoder.num_scales=2 \
model.encoder.upsample_factor=2 \
model.encoder.lowest_feature_resolution=4 \
model.encoder.monodepth_vit_type=vitl \
checkpointing.pretrained_model=pretrained/depthsplat-gs-large-re10k-256x256-view2-e0f0f27a.pth
```

--------------------------------

### Create and Activate Conda Environment

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Sets up a Python 3.10 environment using Conda for the project. This helps manage dependencies.

```bash
# conda
conda create -y -n depthsplat python=3.10
conda activate depthsplat
```

--------------------------------

### Create Symbolic Link for Model Weights

Source: https://github.com/cvg/depthsplat/blob/main/MODEL_ZOO.md

Use this command to create a symbolic link to your downloaded model weights directory. Ensure the path points to your actual model storage location.

```bash
ln -s YOUR_MODEL_PATH pretrained
```

--------------------------------

### Download UniMatch Pre-trained Weights

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Downloads the UniMatch pre-trained weights for model training. Save to the 'pretrained' directory.

```bash
wget https://s3.eu-central-1.amazonaws.com/avg-projects/unimatch/pretrained/gmflow-scale1-things-e9887eda.pth -P pretrained
```

--------------------------------

### DepthSplat Zero-Shot ACID Test (2 Context Views)

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Command for zero-shot generalization from RealEstate10K to ACID, using 2 context views. Ensure pre-trained weights are available.

```bash
# Table 8 of depthsplat paper
CUDA_VISIBLE_DEVICES=0 python -m src.main +experiment=re10k \
mode=test \
dataset.roots=[datasets/acid] \
dataset.view_sampler.index_path=assets/evaluation_index_acid.json \
dataset/view_sampler=evaluation \
dataset.view_sampler.num_context_views=2 \
model.encoder.num_scales=2 \
model.encoder.upsample_factor=2 \
model.encoder.lowest_feature_resolution=4 \
model.encoder.monodepth_vit_type=vitl \
checkpointing.pretrained_model=pretrained/depthsplat-gs-large-re10k-256x256-view2-e0f0f27a.pth
```

--------------------------------

### Render Video on DL3DV (512x960)

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Renders a video on the DL3DV dataset with 12 input views at 512x960 resolution. Requires ffmpeg. Includes camera stabilization option.

```bash
# render video on dl3dv (need to have ffmpeg installed)
CUDA_VISIBLE_DEVICES=0 python -m src.main +experiment=dl3dv \
dataset.test_chunk_interval=1 \
dataset.roots=[datasets/dl3dv_960p] \
dataset.image_shape=[512,960] \
dataset.ori_image_shape=[540,960] \
model.encoder.upsample_factor=8 \
model.encoder.lowest_feature_resolution=8 \
model.encoder.gaussian_adapter.gaussian_scale_max=0.1 \
checkpointing.pretrained_model=pretrained/depthsplat-gs-small-re10kdl3dv-448x768-randview4-10-c08188db.pth \
mode=test \
dataset/view_sampler=evaluation \
dataset.view_sampler.num_context_views=12 \
dataset.view_sampler.index_path=assets/dl3dv_start_0_distance_100_ctx_12v_video.json \
test.save_video=true \
test.stablize_camera=true \
test.compute_scores=false \
test.render_chunk_size=10 \
output_dir=outputs/depthsplat-dl3dv-512x960
```

--------------------------------

### Download DL3DV 960p Test Subset

Source: https://github.com/cvg/depthsplat/blob/main/DATASETS.md

Download a preprocessed subset of the DL3DV 960p dataset for quick inference. This subset contains two test scenes and is for research purposes only.

```bash
wget https://huggingface.co/datasets/haofeixu/depthsplat/resolve/main/dl3dv_960p_test_subset.zip
```

--------------------------------

### Create Symbolic Link to Datasets

Source: https://github.com/cvg/depthsplat/blob/main/DATASETS.md

Use this command to create a symbolic link from your dataset path to the 'datasets' directory, allowing the codebase to access your data.

```bash
ln -s YOUR_DATASET_PATH datasets
```

--------------------------------

### Render Video on re10k (256x256)

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Renders a video on the re10k dataset using 2 input views at 256x256 resolution. Requires ffmpeg. Specifies model encoder and checkpoint.

```bash
# render video on re10k (need to have ffmpeg installed)
CUDA_VISIBLE_DEVICES=0 python -m src.main +experiment=re10k \
dataset.test_chunk_interval=100 \
model.encoder.num_scales=2 \
model.encoder.upsample_factor=2 \
model.encoder.lowest_feature_resolution=4 \
model.encoder.monodepth_vit_type=vitl \
checkpointing.pretrained_model=pretrained/depthsplat-gs-large-re10k-256x256-view2-e0f0f27a.pth \
mode=test \
dataset/view_sampler=evaluation \
dataset.view_sampler.index_path=assets/evaluation_index_re10k_video.json \
test.save_video=true \
test.compute_scores=false
output_dir=outputs/depthsplat-re10k
```

--------------------------------

### DepthSplat DL3DV Test (4 Context Views)

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Command to run DepthSplat in test mode for DL3DV dataset with 4 context views. Ensure pre-trained weights are available.

```bash
# Table 7 of depthsplat paper
CUDA_VISIBLE_DEVICES=0 python -m src.main +experiment=dl3dv \
mode=test \
dataset/view_sampler=evaluation \
dataset.view_sampler.num_context_views=4 \
dataset.view_sampler.index_path=assets/dl3dv_start_0_distance_50_ctx_4v_video_0_50.json \
model.encoder.num_scales=2 \
model.encoder.upsample_factor=4 \
model.encoder.lowest_feature_resolution=8 \
model.encoder.monodepth_vit_type=vitb \
checkpointing.pretrained_model=pretrained/depthsplat-gs-base-dl3dv-256x448-randview2-6-02c7b19d.pth
```

--------------------------------

### DepthSplat DL3DV Test (6 Context Views)

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Command to run DepthSplat in test mode for DL3DV dataset with 6 context views. Ensure pre-trained weights are available.

```bash
# Table 7 of depthsplat paper
CUDA_VISIBLE_DEVICES=0 python -m src.main +experiment=dl3dv \
mode=test \
dataset/view_sampler=evaluation \
dataset.view_sampler.num_context_views=6 \
dataset.view_sampler.index_path=assets/dl3dv_start_0_distance_50_ctx_6v_video_0_50.json \
model.encoder.num_scales=2 \
model.encoder.upsample_factor=4 \
model.encoder.lowest_feature_resolution=8 \
model.encoder.monodepth_vit_type=vitb \
checkpointing.pretrained_model=pretrained/depthsplat-gs-base-dl3dv-256x448-randview2-6-02c7b19d.pth
```

--------------------------------

### Download ACID Test Set

Source: https://github.com/cvg/depthsplat/blob/main/DATASETS.md

Download the ACID test set for evaluation. This is provided by the pixelSplat repo.

```bash
wget http://schadenfreude.csail.mit.edu:8000/re10k_test_only.zip
```

--------------------------------

### DepthSplat DL3DV Test (2 Context Views)

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Command to run DepthSplat in test mode for DL3DV dataset with 2 context views. Ensure pre-trained weights are available.

```bash
# Table 7 of depthsplat paper
CUDA_VISIBLE_DEVICES=0 python -m src.main +experiment=dl3dv \
mode=test \
dataset/view_sampler=evaluation \
dataset.view_sampler.num_context_views=2 \
dataset.view_sampler.index_path=assets/dl3dv_start_0_distance_50_ctx_2v_video_0_50.json \
model.encoder.num_scales=2 \
model.encoder.upsample_factor=4 \
model.encoder.lowest_feature_resolution=8 \
model.encoder.monodepth_vit_type=vitb \
checkpointing.pretrained_model=pretrained/depthsplat-gs-base-dl3dv-256x448-randview2-6-02c7b19d.pth
```

--------------------------------

### Specify Dataset Roots in Configuration

Source: https://github.com/cvg/depthsplat/blob/main/DATASETS.md

Alternatively, you can specify the dataset paths directly in the configuration file using the 'dataset.roots' parameter.

```bash
dataset.roots=[YOUR_DATASET_PATH]/re10k
dataset.roots=[YOUR_DATASET_PATH]/dl3dv
```

--------------------------------

### Evaluate Small Model on RealEstate10K

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Evaluates the small DepthSplat model on the RealEstate10K dataset. This script is used for Table 1 in the paper.

```bash
# Table 1 of depthsplat paper
CUDA_VISIBLE_DEVICES=0 python -m src.main +experiment=re10k \
dataset.test_chunk_interval=1 \
model.encoder.upsample_factor=4 \
model.encoder.lowest_feature_resolution=4 \
checkpointing.pretrained_model=pretrained/depthsplat-gs-small-re10k-256x256-view2-cfeab6b1.pth \
mode=test \
dataset/view_sampler=evaluation
```

--------------------------------

### Evaluate Base Model on RealEstate10K

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Evaluates the base DepthSplat model on the RealEstate10K dataset. This script is used for Table 1 in the paper.

```bash
# Table 1 of depthsplat paper
CUDA_VISIBLE_DEVICES=0 python -m src.main +experiment=re10k \
dataset.test_chunk_interval=1 \
model.encoder.num_scales=2 \
model.encoder.upsample_factor=2 \
model.encoder.lowest_feature_resolution=4 \
model.encoder.monodepth_vit_type=vitb \
checkpointing.pretrained_model=pretrained/depthsplat-gs-base-re10k-256x256-view2-ca7b6795.pth \
mode=test \
dataset/view_sampler=evaluation
```

--------------------------------

### Verify Downloaded File Integrity

Source: https://github.com/cvg/depthsplat/blob/main/MODEL_ZOO.md

Check the integrity of downloaded model files using their sha256sum. This command compares the file's checksum against the prefix in its filename.

```bash
sha256sum filename
```

--------------------------------

### Evaluate Large Model on RealEstate10K

Source: https://github.com/cvg/depthsplat/blob/main/README.md

Evaluates the large DepthSplat model on the RealEstate10K dataset. This script is used for Table 1 in the paper.

```bash
# Table 1 of depthsplat paper
CUDA_VISIBLE_DEVICES=0 python -m src.main +experiment=re10k \
dataset.test_chunk_interval=1 \
model.encoder.num_scales=2 \
model.encoder.upsample_factor=2 \
model.encoder.lowest_feature_resolution=4 \
model.encoder.monodepth_vit_type=vitl \
checkpointing.pretrained_model=pretrained/depthsplat-gs-large-re10k-256x256-view2-e0f0f27a.pth \
mode=test \
dataset/view_sampler=evaluation
```

--------------------------------

### Convert DL3DV Test Set

Source: https://github.com/cvg/depthsplat/blob/main/DATASETS.md

Use this script to convert the DL3DV test set before generating the index file. Ensure dataset paths are updated.

```bash
python src/scripts/convert_dl3dv_test.py
```

--------------------------------

### Modify Python Script for 720p Download

Source: https://github.com/cvg/depthsplat/blob/main/DATASETS.md

When downloading the 720p RealEstate10K dataset, modify the specified line in the Python script to change '360p' to '720p'.

```python
360p
```

--------------------------------

### Convert DL3DV Training Set

Source: https://github.com/cvg/depthsplat/blob/main/DATASETS.md

Use this script to convert the DL3DV training set. Excludes scenes present in the test set. Ensure dataset paths are updated.

```bash
python src/scripts/convert_dl3dv_train.py
```

--------------------------------

### Generate DL3DV Test Index

Source: https://github.com/cvg/depthsplat/blob/main/DATASETS.md

Generates the index.json file for the DL3DV test set after conversion.

```bash
python src/scripts/generate_dl3dv_index.py
```

--------------------------------

### Process High-Resolution DL3DV Data

Source: https://github.com/cvg/depthsplat/blob/main/DATASETS.md

Follow the same procedure as the standard DL3DV dataset processing, but update the 'images_8' folder to 'images_4' for high-resolution data.

```bash
# Example command structure, actual script may vary
# python src/scripts/convert_dl3dv_highres.py --image_folder images_4
```

--------------------------------

### DepthSplat Citation

Source: https://github.com/cvg/depthsplat/blob/main/README.md

This is the citation for the DepthSplat project in BibTeX format. Include this in your academic publications when referencing the work.

```bibtex
@inproceedings{xu2024depthsplat,
      title   = {DepthSplat: Connecting Gaussian Splatting and Depth},
      author  = {Xu, Haofei and Peng, Songyou and Wang, Fangjinhua and Blum, Hermann and Barath, Daniel and Geiger, Andreas and Pollefeys, Marc},
      booktitle={CVPR},
      year={2025}
    }
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.