Object tracking is a computer vision task used to follow the movement of objects across video
frames, helping systems monitor and understand how things change over time. It is used in
a wide range of real-world scenarios from estimating speed and direction of moving objects to
counting objects in videos.
Systems integrated with object tracking can be configured to trigger real-time alerts when
something unusual is detected, such as a person entering a restricted area or a delivery being
left too long in one place.
Instance segmentation and tracking
Instance segmentation is an advanced computer vision task that combines object detection with
pixel-level segmentation. It identifies and outlines individual objects within an image or video,
providing detailed masks that delineate the boundaries of each object. This is particularly useful in
applications where precise object boundaries are important, such as in autonomous driving, robotics,
and medical imaging. By providing a richer representation of the visual world, instance segmentation
enables tasks like object counting, tracking, and interaction analysis.
It is useful in cases where you need to measure the size of detected objects, like in drone image
analysis. However, instance segmentation models are typically larger, slower, and less accurate than
object detection models. Moreover, they require more data to train effectively, and the annotation
process is significantly more laborious, with a harder training process. So, it is often better to try
object detection first because it is easier and cheaper to test, and faster and more accurate to run.
Instance segmentation only must be used when the outline of the object is required by the application.
Otherwise, object detection is enough for most cases.
How to train an instance segmentation model with DeepLab
Custom training is the process of fine-tunning a pre-trained model on a specific dataset so it can
recognize objects that are not included in the original training dataset.
To train an instance segmentation model with DeepLab, you need to follow these steps:
Prepare your dataset: create a new project folder and upload your images, then use the auto segmentation
tool to annotate the objects using an outline polygon.
Train the model: once the dataset has been fully annotated, just click "train" to launch the notebook
and start the training process. Just select the dataset folder and start training your model.
Run the model: once the training is finished, you can run the model on your images or videos using the
DeepLab app. Just select the input file and model and and click "run" to detect and segment the objects in
the input file.
