Similari VS sort

Hello, community. I have released a new version of Similari - 0.26.2.

Well, I never used the original DeepSORT implementation from the repo, but as I know you can pass any ReID to DeepSORT's predict if the metric is compatible. E.g. I have my own tracking library with DeepSORT flavor implemented. To run predict one just fill the structure like demonstrated here: https://github.com/insight-platform/Similari/blob/main/python/visual_sort.py

Benchmark results;

Both trackers support axis-aligned and rotated bounding boxes and work really fast (4-10 times faster than NumPy SORT depending on the number of objects in the scene).

Rust code (Similari framework repo): https://github.com/insight-platform/Similari/blob/main/src/utils/kalman.rs

My work is broadly connected with object trackers development, recently I have had a chance to publish an open source framework for developing object trackers. The framework name is Similari, it is written in Rust, making it a high-performance and robust solution. Feature vector operations are optimized for SIMD. I wrote an introductory article on Medium that explains how to build an IoU tracker with Similari: article. Also, the project repository contains a set of examples of how to construct trackers.

Several days ago, I posted the announcement about Similari - a rust framework for building tracker and similarity search engines. That post was positively accepted, but a couple of people wanted to learn more about the applicability of Similari. To fill the gap, I wrote an article on Medium where I demonstrated how to build a simple IoU tracker with Similari. I also do my best to fill the repo with useful samples - you can look at the samples directory to find how to build various trackers with the framework.

you can construct simple vectorwise searches, see an example, it's not as performant as HNSW but more universal;

Similari is a Rust/Python framework aimed at building sophisticated tracking systems. With Similari, you can develop highly efficient parallelized SORT, DeepSORT, and other sophisticated multiple-object tracking engines.

One interesting thing to consider is about using a new model's output as training data is the following. If you use some object tracking algorithm like SORT, you might be able to find frames where the object was "tracked" even though the model didn't detect the object (some type of flickering). In this case, you're able to run the model on video and make it better by giving it these new bounding boxes based on "tracking". Does that make sense?

The model is an ST-GCN trained in a completely unsupervised manner, from a LOT of skeleton sequences. Pose estimation was performed with AlphaPose, and tracked using SORT.

Yes it is possible! It's an addition to object detection, so I suggest you look for "object detection tracking". You can find articles like this one which refers to Python libraries like this one.

congratz! that's a great project. since you have the detector operating on videos, you could add a simple tracking-by-detection method like SORT. https://github.com/abewley/sort#using-sort-in-your-own-project it's really easy to use, and creates temporally constistent tracks by associating detections in successive frames.