# State of the art object tracking with neural networks

**URL:** <https://discourse.julialang.org/t/state-of-the-art-object-tracking-with-neural-networks/101710>\
**Category:** Machine Learning\
**Tags:** neural-network, literature\
**Created:** [July 17, 2023, 10:16pm UTC](https://discourse.julialang.org/t/state-of-the-art-object-tracking-with-neural-networks/101710 "2023-07-17T22:16:50Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![juliohm](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/juliohm/32/215266_2.png) [@juliohm](https://discourse.julialang.org/u/juliohm)\
**Post date:** [July 17, 2023, 10:16pm UTC](https://discourse.julialang.org/t/state-of-the-art-object-tracking-with-neural-networks/101710/1 "2023-07-17T22:16:50Z")

</div>

Anyone following the latest models for object tracking in computer vision?

Object tracking consists of identifying an object (or bounding box of the object) on every single frame of a video. It usually combines object detection with motion prediction to handle occlusion and other subtle issues that cannot be solved with object detection alone.

I found a couple of models like Detectron and Detectron2 by Facebook research, but it would be nice to get additional resources from people working actively in this field.

---

<div class="post-metadata">

**Author:** ![Ashwani\_Rathee](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/ashwani_rathee/32/49551_2.png) [@Ashwani\_Rathee](https://discourse.julialang.org/u/Ashwani_Rathee)\
**Post date:** [July 18, 2023, 5:24am UTC](https://discourse.julialang.org/t/state-of-the-art-object-tracking-with-neural-networks/101710/2 "2023-07-18T05:24:36Z")

</div>

After next 10th, I am also interested in this for JuliaImages and all downstream use cases that can serve many fellow organisations.

Might be of interest(mostly classical though)

- [GitHub - baggepinnen/BlobTracking.jl: Detect and track blobs in video](https://github.com/baggepinnen/BlobTracking.jl)
- [GitHub - JuliaImages/ImageTracking.jl](https://github.com/JuliaImages/ImageTracking.jl)
- [GitHub - r3tex/ObjectDetector.jl: Pure Julia implementations of single-pass object detection neural networks.](https://github.com/r3tex/ObjectDetector.jl) with a kalman filter might make for fine object trackers too.

@yakir12 is also interested in this for his work.

---

<div class="post-metadata">

**Author:** ![DoktorMike](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/doktormike/32/2736_2.png) [@DoktorMike](https://discourse.julialang.org/u/DoktorMike)\
**Post date:** [July 18, 2023, 7:47am UTC](https://discourse.julialang.org/t/state-of-the-art-object-tracking-with-neural-networks/101710/3 "2023-07-18T07:47:43Z")

</div>

I believe that liquid neural networks are the SOTA currently for this. I also think it’s one of the best approaches for this currently.

> **[Drones navigate unseen environments with liquid neural networks](https://news.mit.edu/2023/drones-navigate-unseen-environments-liquid-neural-networks-0419)**
>
> MIT CSAIL researchers have used liquid neural networks to enable vision-based autonomous fly-to-target tasks under distribution shifts.

This looks at navigation but have in earlier papers focused on realtime object tracking.

Just my two cents.

---

<div class="post-metadata">

**Author:** ![Jay-sanjay](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/jay-sanjay/32/52128_2.png) [@Jay-sanjay](https://discourse.julialang.org/u/Jay-sanjay)\
**Post date:** [August 9, 2023, 7:04am UTC](https://discourse.julialang.org/t/state-of-the-art-object-tracking-with-neural-networks/101710/4 "2023-08-09T07:04:45Z")

</div>

The Liquid Neural Networks are the best for it as they have the distinctive architecture that holds potential advantages for various computer vision tasks. LNNs are specialized for handling temporal data with long-range dependencies and complex patterns. They would excel in scenarios where the visual dynamics are intricate, evolving over longer time intervals, or where understanding complex interactions is essential. RNNs, on the other hand, are versatile for a wide range of sequential data tasks and may be more suitable for simpler temporal relationships in visual data.  
So LLNs are a better choice…