I’m currently engaged in PhD research focused on multiple object tracking. My team and I are actively utilizing Labelbox for our project. Recently, we noticed that Labelbox has introduced an automatic tracking function. While we appreciate this new feature and understand its potential benefits, it unfortunately does not meet our specific needs and requirements.
Given our unique challenges, we have developed a custom tracking algorithm which we believe is better suited to our project (Something like trackanything trained on custom data). However, we’re encountering a hurdle with integrating this algorithm into the existing Labelbox framework. Our main question is: Is it possible for us to override the current automatic tracking method in Labelbox with our custom-designed tracking algorithm?
We are completely prepared to devote substantial time to any engineering tasks necessary for this integration. We understand that these tasks may be complex and are willing to work through them.
Hi! Thanks for your message. I am Jay Patel, Senior PM for Computer Vision at Labelbox.
At the moment, the tracking model on our video editor is not “swappable” with something else. We have something on the backend that we make requests to, but it’s on our end so it’s not accessible from the SDK.
Are you open to scheduling a call so we can communicate about this better? If so, please share your email and some times that work for you for a 30 min chat.
While it’s unfortunate that the tracking model is not readily swappable, I appreciate your willingness to discuss this further.
My email address is atom.james.scott@gmail.com. Please find my availability for a 30-minute call on the following dates and times (all in Pacific Time / GMT-7):
Thursday, June 30th: 4:00pm - 10:00pm
Sunday, July 3rd: 4:00pm - 10:00pm
Monday, July 4th: 4:00pm - 10:00pm
Apologies for the late hours due to the time difference, given my current location in Japan. Feel free to email me to confirm a suitable slot.