Python scripts form performing stereo depth estimation using the HITNET model in Tensorflow Lite.
Stereo depth estimation on the cones images from the Middlebury dataset (https://vision.middlebury.edu/stereo/data/scenes2003/)
- OpenCV, imread-from-url and tensorflow==2.6.0 or tflite_runtime. Also, pafy and youtube-dl are required for youtube video inference.
pip install -r requirements.txt
pip install pafy youtube-dl
For the tflite runtime, you can either use tensorflow(make sure it is version 2.6.0 or above) pip install tensorflow==2.6.0 or the TensorFlow Runtime binary
In computers with a GPU, the program would silently creash without any error during the inference, os.environ["CUDA_VISIBLE_DEVICES"]="-1" is added at the beginning of the script to force the program to run on the CPU. You can comment this line for other types of devices.
The original models were converted to different formats (including .tflite) by PINTO0309, download the models from his repository and save them into the models folder.
The Tensorflow pretrained model was taken from the original repository.
- Image inference:
python imageDepthEstimation.py
- Video inference:
python videoDepthEstimation.py
- DrivingStereo dataset inference:
python drivingStereoTest.py
For performing the inference in Tensorflow, check my other repository HITNET Stereo Depth estimation.
For performing the inference in ONNX, check my other repository ONNX HITNET Stereo Depth estimation.
- Hitnet model: https://github.com/google-research/google-research/tree/master/hitnet
- PINTO0309's model zoo: https://github.com/PINTO0309/PINTO_model_zoo
- PINTO0309's model conversion tool: https://github.com/PINTO0309/openvino2tensorflow
- DrivingStereo dataset: https://drivingstereo-dataset.github.io/
- Original paper: https://arxiv.org/abs/2007.12140