DAGE: Dual-Stream Architecture for Efficient and Fine-Grained Geometry Estimation

DAGE delivers accurate and consistent 3D geometry, fine-grained and high-resolution depthmaps, while maintaining efficiency and scalability.

DAGE is a dual-stream transformer that disentangles global coherence from fine detail for geometry estimation from uncalibrated multi-view/video inputs.

• LR stream builds view-consistent representations and estimates cameras efficiently.
• HR stream preserves sharp boundaries and fine structures per-frame.
• Lightweight adapter fuses the two via cross-attention without disturbing the pretrained single-frame pathway.
• Scales resolution and clip length independently, supports inputs up to 2K, and achieves state-of-the-art on video geometry estimation and multi-view reconstruction.

• [Mar, 2026] Initial release with inference, training code and model checkpoint.

git clone https://github.com/ngoductuanlhp/DAGE.git
cd DAGE

bash scripts/instal_env.sh
conda activate dage

This creates a conda environment with Python 3.10, PyTorch 2.10.0 (CUDA 13.0), and all required dependencies.

Run on the included demo data or your own video/image folder:

# Run with default settings on demo data
bash demo.sh

# Or run directly with custom arguments

# Default: LR at 252px, HR at 3600 tokens (~840x840 for square images)
python inference/infer_dage.py --checkpoint TuanNgo/DAGE

# Higher LR resolution (better camera poses, more compute)
python inference/infer_dage.py --checkpoint TuanNgo/DAGE --lr_max_size 518

# Higher HR resolution up to 2K (sharper pointmaps)
python inference/infer_dage.py --checkpoint TuanNgo/DAGE --hr_max_size 1920

# Memory-efficient chunking for GPUs with <40GB VRAM (lower chunk_size if OOM)
python inference/infer_dage.py --checkpoint TuanNgo/DAGE --hr_max_size 1920 --chunk_size 8

Arguments:

Input: Place videos (.mp4, .MOV) or image folders in assets/demo_data/.

Output: For each input, the script saves:

• <name>_disp_colored.mp4 — colorized disparity video
• <name>_depth_colored.mp4 — colorized depth video
• <name>.npy — dictionary with pointmap, pointmap_global, pointmap_mask, rgb, and extrinsics

Our checkpoint is available at 🤗 Hugging Face Hub: TuanNgo/DAGE

Or you can manually download the checkpoint and place it in the checkpoints/ directory:

mkdir -p checkpoints

gdown --fuzzy https://drive.google.com/file/d/1BsBJ7MTarlBP5RjCVfPQoQMsCxccBabF/view?usp=sharing -O ./checkpoints/

• Input: torch.Tensor of shape (B, N, 3, H, W) with pixel values in [0, 1].
• Output: A dict with the following keys:

import torch
from einops import rearrange

from dage.models.dage import DAGE
from dage.utils.data_utils import read_video, resize_to_max_side

# --- Setup ---
device = 'cuda'
model = DAGE.from_pretrained('checkpoints/model.pt').to(device).eval()

# --- Load Data ---
# read_video returns (frames, H, W, fps)
# Options: stride=N, max_frames=N, force_num_frames=N
video, H, W, fps = read_video('path/to/video.mp4', stride=10, max_frames=100)

# Prepare tensors (B, N, C, H, W), values in [0, 1]

lr_max_size = 252
hr_max_size = 518 # or 1022 / 1918

lr_video, lr_height, lr_width = resize_to_max_side(video, lr_max_size)
hr_video, hr_height, hr_width = resize_to_max_side(video, hr_max_size)  
hr_num_tokens = (hr_height // 14) * (hr_width // 14)

lr_video = rearrange(torch.from_numpy(lr_video), 't h w c -> 1 t c h w').float().to(device) / 255.0
hr_video = rearrange(torch.from_numpy(hr_video), 't h w c -> 1 t c h w').float().to(device) / 255.0

# --- Inference ---
with torch.no_grad():
    output = model.infer(
        hr_video=hr_video,
        lr_video=lr_video,
        lr_max_size=lr_max_size,
        hr_num_tokens=hr_num_tokens,
        chunk_size=None,  # optional, for memory efficiency
    )

# Access outputs
local_points = output['local_points']   # (N, H, W, 3)
global_points = output['global_points'] # (N, H, W, 3)
camera_poses = output['camera_poses']   # (N, 4, 4)
mask = output['mask']                   # (N, H, W)

Both streams require resolutions that are multiples of the patch size (14). The HR stream defaults to 3600 tokens total (e.g., 840x840 for square images, 630x1120 for 9:16), but can be overridden with --hr_max_size.

We use viser for interactive 3D point cloud visualization. The inference script saves .npy files that can be directly visualized.

Dynamic scenes — renders pointmaps sequentially with playback controls:

python visualization/vis_pointmaps.py --data_path quali_results/dage/<name>.npy

# NOTE removing floating points at edges (if exist)
# python visualization/vis_pointmaps.py --data_path quali_results/dage/<name>.npy --filter_edge

Static scenes — merges all frames into a single point cloud in a shared coordinate frame:

python visualization/vis_pointmaps_all.py --data_path quali_results/dage/<name>.npy

# NOTE removing floating points at edges (if exist)
# python visualization/vis_pointmaps_all.py --data_path quali_results/dage/<name>.npy --filter_edge

See docs/TRAINING.md for detailed instructions on data preparation, loss functions, and configuration.

See docs/EVALUATION.md for detailed instructions.

DAGE/
├── assets/
│   └── demo_data/                  # Demo videos for inference
├── configs/
│   └── model_config_dage.yaml      # Model architecture config
├── dage/                           # Main package
│   ├── models/
│   │   ├── dage.py                 # DAGE model
│   │   ├── dinov2/                 # DINOv2 backbone
│   │   ├── layers/                 # Transformer blocks, attention, camera head
│   │   └── moge/                   # MoGe encoder components
│   └── utils/                      # Geometry, visualization, data loading
├── evaluation/                     # Benchmark evaluation
├── inference/
│   └── infer_dage.py               # Main inference script
├── scripts/
│   ├── eval/                       # Evaluation bash scripts
│   ├── infer/                      # Inference bash scripts
│   └── instal_env.sh               # Environment setup
├── setup.py
├── third_party/                    # Code for related work (VGGT, Pi3, Cut3r, etc)
└── training/
    ├── dataloaders/                # Video dataloaders & dataset configs
    ├── loss/                       # Loss functions
    ├── train_dage_stage{1,2,3}.py  # Three-stage training scripts
    └── training_configs/           # YAML configs for trainings

Our work builds upon several open-source projects:

• DUSt3R
• Pi3
• MoGe
• VGGT
• DINOv2

If you find our work useful, please consider citing:

@inproceedings{ngo2026dage,
  title={DAGE: Dual-Stream Architecture for Efficient and Fine-Grained Geometry Estimation},
  author={Ngo, Tuan Duc and Huang, Jiahui and Oh, Seoung Wug and Blackburn-Matzen, Kevin and Kalogerakis, Evangelos and Gan, Chuang and Lee, Joon-Young},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2026}
}

The code in this repository is released under the CC BY-NC 4.0 license, unless otherwise specified.