Instance segmentation and tracking system in realtime
Budget: €1,500 – €3,000 EUR
Instance segmentation + tracking system need to be built.
We are looking for someone we can build real-time video instance segmentation + tracking system.
Input: video file/rtmp stream whatever is easy for you.
Output: instance segmentation masks, bounding boxes, confident score, tracking id, object name
Phase 1: instance segmentation, bounding box, confident score, object name
Phase 2: phase 1 result + tracking id
Phase 3: compressed metadata.
DNNs are normally computationally heavy, we want to run DNNs on a dedicated GPU server and return results to client application. Due to network speed considerations, we want to represent the metadata in a compressed format. Bounding boxes, confident score tracking id does not take too much size but instance segmentation mask. We are primarily thinking of compressed polygon representation for instance masks but new ideas are welcome as long as it meets our goals.
Anyone who can demonstrate any of the subtasks are welcome. For example, the candidate can work on Phase3 assuming a instance segmentation ask in single channel image format where each pixel represent a tracking id or object class.
We are looking for someone we can build real-time video instance segmentation + tracking system.
Input: video file/rtmp stream whatever is easy for you.
Output: instance segmentation masks, bounding boxes, confident score, tracking id, object name
Phase 1: instance segmentation, bounding box, confident score, object name
Phase 2: phase 1 result + tracking id
Phase 3: compressed metadata.
DNNs are normally computationally heavy, we want to run DNNs on a dedicated GPU server and return results to client application. Due to network speed considerations, we want to represent the metadata in a compressed format. Bounding boxes, confident score tracking id does not take too much size but instance segmentation mask. We are primarily thinking of compressed polygon representation for instance masks but new ideas are welcome as long as it meets our goals.
Anyone who can demonstrate any of the subtasks are welcome. For example, the candidate can work on Phase3 assuming a instance segmentation ask in single channel image format where each pixel represent a tracking id or object class.