Practical overview
When it fits
Use Densedepth to inspect monocular image-to-depth inference or as a codebase for dataset-based training and evaluation. The pretrained model path is the documented starting point for inference.
Who it is for
- Researchers and developers exploring monocular depth estimation.
- Practitioners who want to generate depth maps from images using pretrained weights.
- Experimenters training or evaluating the model with NYU Depth V2 or KITTI data.
Documented uses
- Estimate depth maps from images with pretrained model weights.
- Visualize point-cloud output from a webcam or image using the Qt demo.
- Train or evaluate the implementation using the documented datasets and scripts.
Requirements & limits
The repository is public and identified as open source under the GPL-3.0 license.
- The primary implementation is Keras with TensorFlow; TensorFlow 2.0 and PyTorch versions are identified as experimental.
- The README's tested setup uses Keras 2.2.4, TensorFlow 1.13, CUDA 10.0 and Windows 10 or Ubuntu 16.
- Inference has tested minimum GPU examples of an NVIDIA GeForce 940MX laptop GPU or GTX 950 desktop GPU.
- KITTI training data must be converted to dense depth maps through inpainting of Lidar depth data.
- The README estimates about 24 hours of training on one NVIDIA TITAN RTX at batch size 8.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Run image inference with the NYU pretrained model
Produce a montage showing input images alongside estimated depth maps.
Before you start
- Download the NYU pretrained model, listed as a 165 MB file.
- Use the repository's Keras and TensorFlow implementation with its required Python packages.
- The README's tested setup uses Keras 2.2.4, TensorFlow 1.13, CUDA 10.0, Windows 10 or Ubuntu 16, and an NVIDIA GPU.
Steps
Download the NYU pretrained model and save it as nyu.h5.
Run the command below.
python test.py
Expected outcome
The README says the script should display a montage of images with estimated depth maps.
Guide limitations
- The README specifies tested software and hardware configurations but does not guarantee performance on other setups.
- The listed minimum hardware tested for inference is an NVIDIA GeForce 940MX laptop GPU or GTX 950 desktop GPU.
Prepare KITTI data for training
Prepare KITTI inputs in the dense-depth format expected by the training workflow.
Before you start
- KITTI raw data.
- A depth-inpainting method for converting Lidar depth data into dense input depth maps.
Steps
Copy the KITTI raw data to a folder at
../kitti.Apply a depth-inpainting method to the Lidar data to produce dense depth maps.
Expected outcome
KITTI data is arranged at the documented path and has the dense depth maps the method expects.
Guide limitations
- This guide covers the documented data-preparation stages only; it does not cover running training.
- The repository's training experiments use a specified KITTI subset.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 16 October 2026
- Original website title
- ialhashim/DenseDepth · Official homepage title · retained in its source language