Visual object tracking is the backbone of diverse subfields inside computer vision, including robot vision and autonomous driving. This job goals to reliably discover the goal object in a video sequence. Many state-of-the-art algorithms compete within the Visual Object Tracking (VOT) challenge because it is one of the crucial essential competitions within the tracking field.

The Visual Object Tracking and Segmentation competition (VOTS2023) removes a few of the restrictions imposed by previous VOT challenges in order that participants can take into consideration object tracking more broadly. In consequence, VOTS2023 combines short- and long-term monitoring of a single goal and tracking many targets, using goal segmentation because the only position specification. This introduces recent difficulties, resembling precise mask estimate, multi-target trajectory tracking, and recognizing relationships between objects.

A brand new study by the Dalian University of Technology, China, and DAMO Academy, Alibaba Group, presents a system called HQTrack, which stands for High-Quality Tracking. It comprises primarily a video multi-object segmenter (VMOS) and a mask refiner (MR). To perceive tiny objects in intricate setups, the researchers employ VMOS, an enhanced variation of DeAOT, and cascade a gated propagation module (GPM) at 1/8 scale. As well as, they use Intern-T as their feature extractor to enhance the power to differentiate between several types of objects. In VMOS, the researchers only keep essentially the most recently used frame within the long-term memory, discarding the older ones to make room. Nonetheless, applying a big segmentation model to enhance the tracking masks could possibly be useful. Objects with complicated structures are especially difficult for SAM to predict, they usually appear often within the VOTS challenge. 

Using an HQ-SAM model that has already been pre-trained, the team may further enhance the standard of the tracking masks. Final tracking results were chosen from VMOS and MR, they usually used the outer enclosing boxes of the expected masks as box prompts to feed into HQ-SAM alongside the unique images to acquire the refined masks. HQTrack finishes in second place on the VOTS2023 competition with a top quality rating of 0.615 on the test set. 

Try the Paper and GitHub. All Credit For This Research Goes To the Researchers on This Project. Also, don’t forget to hitch our 27k+ ML SubReddit, Discord Channel, and Email Newsletter, where we share the most recent AI research news, cool AI projects, and more.