Official implementation of of BRIDGE: Building Reinforcement-Learning Depth-to-Image Data Generation Engine for Monocular Depth Estimation

• [2025-09-30] 🚀🚀🚀 We published BRIDGE on arXiv and demos on huggingface! Try our DEMO.
• [2025-09-30] 🎉🎉🎉 We released the model checkpoint on huggingface.

1. We present BRIDGE, an RL-optimized, large-scale Depth-to-Image (D2I) data engine. It generates massive, high-quality RGB-D data to address critical Monocular Depth Estimation (MDE) training challenges and foster robust real-world performance.

   Our main contributions are summarized as follows:

   • An efficient RL-driven D2I data engine: BRIDGE efficiently generates over 20 million diverse, high-quality RGB-D data from synthetic depth, alleviating data scarcity and quality issues.
   • A novel hybrid depth supervision strategy: We introduce a hybrid training strategy combining generated RGB with high-precision ground truth and teacher pseudo-labels, enhancing geometric knowledge learning.
   • Superior performance and high training efficiency: BRIDGE achieves SOTA MDE performance across benchmarks with significantly less data (20M vs. 62M), demonstrating excellent detail capture and robustness.

Download the checkpoint from huggingface and put it under the checkpoints directory.

git clone https://github.com/lnbxldn/Bridge.git
cd Bridge
pip install -r requirements.txt

import cv2
import torch
import numpy as np
from bridge.dpt import Bridge 
model = Bridge()
model.load_state_dict(torch.load(f'checkpoints/bridge.pth', map_location='cpu'))
model = model.to(DEVICE).eval()

raw_img = cv2.imread('your/image/path')
depth = model.infer_image(raw_img)  

If you find this project useful, please citing:

@misc{Liu2025BRIDGE,
  title={BRIDGE - Building Reinforcement-Learning Depth-to-Image Data Generation Engine for Monocular Depth Estimation},
  author={Liu, Dingning and Guo, Haoyu and Zhou, Jingyi and He, Tong},
  year={2025},
  eprint={2509.25077},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2509.25077},
}