Contribution 6: Crowd Counting From Single View with SSDCNet
I have successfully completed the first milestone of my GSoC which was to count people present in the scene given a video feed using existing deep Net models. In my earlier post, I had mentioned that I will be going foreward with the SS-DCNet model due the advantages which I had discussed. I have used the models that were trained the Shanghai Dataset(A, B) and QRNF dataset by the authors of the SS-DCNet paper(ICCV 2019). This is the link to my repository. The inputs are one of the three pretrained models and a video file(which may be a URL or a file stored in the videos folder of the repository). In the event that a video is not not supplied a reference to the webcam will be grabbed.
An example of the of the People Counter employed through SS-DCNet
Observations:
The SS-DCNet fails sometimes when the number of people in the scene is say 2 or 3 as it is trained on overcrowded datasets. This may resolved if we train the network to perform in the setting in which we want to count the people or have a model that is suited to scarcely populated images whenever the SS-DCNet returns a count less than some given threshold. Another possible solution is to have a some filtering techniques to smooth out sudden jumps which seem unrealistic in a continuous video feed. This I will try to resolve in the coming weeks on the advise of my mentors.