The project aims to implement a depth estimation system with a single camera, implemented with Keras.
To have an overview of the project, check the poster in the repository
-
Make sure you have python development environment on your PC.
-
Before running the training and testing scripts, first run below command to install dependencies
pip install -r "requirements.txt"
This project is tested with Keras 2.4.3 (tensorflow 2.3.0 as backend), CUDA 11.0, Python 3.6.7 on Ubuntu 18.04 and Windows 10
-
Download Training & Testing Set and Trained Model to the root directory (no need to extract) and corresponding folder
- The model is trained on the platform with NVIDIA Tesla K80 GPU @ 240.48 GB/s 12GB, at the bs = 4
- Run
python train.py
, see more helping info with adding-h
as an extra argument in CMD
- To evaluate the model run
python evaluate.py
, add--model
to import your own trained model - You will have the six evaluation criteria (threshold accuracies, average relative error, root mean squared error, average error):
- Run
python predict.py --input yourpic.jpg
- Note: if you download the trained model, the detection range is limited to 10 meters
The project will be used in "Flea Market" project for indoor prediction and 3D reconstruction.