During my research I usually like to visuallize and understand clearly how some papers/models work. In this repository I will create some tutorials in Colab for repositories that I found dificult to understand how to use from their given code.
This approach consists on detecting objects in videos with a given identifier that is maintained during the whole video.
- Multi-Object Tracking Detection Tutorial: use of tracking-detection approach to detecting objects from images and videos and tracking their position through time. We make use of the most state-of-the-art obejct detectors with a real-time tracer for easy deployment to real world scenarios
In order to really understand the context of a given video, it is not only necessary to track and detect people, but to be able to undersatnd how these people interacts with the objects of the scene. Human-Object Interaction (HOI) detect this pairs of relationships between humans and objects and classifies the given task.
- QPIC Tutorial: original repository is Query-Based Pairwise Human-Object Interaction Detection with Image-Wide Contextual Information
- CDN Tutorial: original repository is Mining the Benefits of Two-stage and One-stage HOI Detection
Detect what people is looking at allows to gain a lot of information regarding the scene and to predict intentions of humans.
- Gaze-Following Tutorial: original repository is Detecting Attended Visual Targets in Video for predicting where humans focuses and Yolov5_DeepSort_Pytorch for the tracking-detection of the faces through time.