Count vehicle using density map regression.
pip3 install requirements.txt
Download the dataset from the link down below. After that, extract the datasets in the project folder. Each car in the frames is annotated by a bounding box.
CityCam dataset
Note that CityCam contains some bad-formed xml files, in which the annotator failed to escape '&'. I also found that in one specific image frame the 'y_min' of the bounding box is out of the image border. All of invalid data I found has been listed in the CityCam_invalid_frames.txt.
Gernally speaking, I put a gaussian kernel at the center of every car to generate a density map for a specific image frame. As for the standard deviation (sigma) of the gaussian kernel, it is determined by the sum of the distances to the k nearest neighbors. Please read this MCNN paper for more details. You could find my implementation over here k_nearest_gaussian_kernel.
To generate the density maps, just type python3 counting_datasets/CityCam_maker.py in the project directory.
First of all, set up your hyper parameters in hyper_param_conf.py. And then, execute the main.py script python3 main.py. The trained model will be stored in checkpoints/, logs will be saved in logs/ for your reference.